DiscoStream_Legacy.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/WASasquatch/discostream/blob/main/DiscoStream_Legacy.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TitleTop"
      },
      "source": [
        "# <font size=\"6\">**WAS**'s DiscoStream *Legacy* ![visitors](https://visitor-badge.glitch.me/badge?page_id=Portrait-Generator-Playground&left_color=blue&right_color=orange) [![GitHub](https://img.shields.io/badge/github-%23121011.svg?style=for-the-badge&logo=github&logoColor=white)](https://github.com/WASasquatch/disco-diffusion-portrait-playground/)</font>\n",
        "\n",
        "<hr>\n",
        "\n",
        "<font size=\"5\">Based on Disco Diffusion</font><br><font size=\"4\"><i>Formerly WAS's Disco Diffusion v5.6-8 - Portrait Playground</i></font>\n",
        "\n",
        "<font size=\"4.5\">In the future, AI Art will become more and more popular as people become more interested in creating and viewing art that is created by machines. This type of art is already beginning to gain popularity and as technology advances, it is likely that AI Art will become even more widespread.\n",
        "\n",
        "If you're interested in art and want to try something new, then you should definitely give AI Art a try. It's a fun and unique way to create art, and you may be surprised at how much you enjoy it.\n",
        "\n",
        "Making text-to-image AI Art is actually quite easy, and anyone can do it with the right software, such as [Disco Diffusion](http://discodiffusion.com/). All you need is a computer with an internet connection and a bit of creativity.\n",
        "\n",
        "Some people use text-2-image art generation to create images of famous quotes or to create images that look like they were hand-drawn.\n",
        "\n",
        "Come and create something new and exciting with us!</font>\n",
        "\n",
        "<font size=\"2\" color=\"9f79b2\">This introduction was generated by Artificial Intelligence as an expresion of the future of helpful tools driven by AI.</font>\n",
        "\n",
        "<br>\n",
        "<hr>\n",
        "<br>\n",
        "\n",
        "#### This Colab is maintained by **WAS**asquatch <font size=\"2\">(<font color=\"7e87f6\">Discord</font>: WAS\\#0263) and **plambe** <font size=\"2\">(<font color=\"7e87f6\">Discord</font>: plambe\\#0263) </font>\n",
        "<br> \n",
        "\n",
        "## **Other Tools:**\n",
        "\n",
        "- [OpenAI Guided Diffusion - Model Training / Fine-Tuning](https://rebrand.ly/guided-diffusion) - A implementation of OpenAI's Guided Diffusion in Google Colabs\n",
        "- [Noodle Soup Prompts](https://rebrand.ly/noodle-soup-prompts) - Terminology Database for prompt exploration\n",
        "- [Portrait Generator `init_image` Prep](https://rebrand.ly/dd-image-prep) - Prep portraits other other `init_images` as approximations for better DD improvisation. \n",
        "\n",
        "<br>\n",
        "\n",
        "-----\n",
        "\n",
        "<br> \n",
        "\n",
        "### Disco Diffusion - http://discodiffusion.com/ , https://github.com/alembics/disco-diffusion - [LICENSE](https://github.com/alembics/disco-diffusion/blob/main/LICENSE)\n",
        "\n",
        "In case of confusion, Disco is the name of this notebook edit. The diffusion model in use is Katherine Crowson's fine-tuned 512x512 model\n",
        "\n",
        "For issues, join the [Disco Diffusion Discord](https://discord.gg/msEZBy4HxA) or message us on twitter at [@somnai_dreams](https://twitter.com/somnai_dreams) or [@gandamu_ml](https://twitter.com/gandamu_ml)\n",
        "\n",
        "<br>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TutorialTop"
      },
      "source": [
        "# <a name=\"reference_guide\">Reference</a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DiffusionSet"
      },
      "source": [
        "**Diffusion Settings Reference**\n",
        "---\n",
        "Disco Diffusion is complex, and continually evolving with new features.  The most current documentation on on Disco Diffusion settings can be found in the unofficial guidebook:\n",
        "\n",
        "[Zippy's Disco Diffusion Cheatsheet](https://docs.google.com/document/d/1l8s7uS2dGqjztYSjPpzlmXLjl5PM3IGkRWI3IiCuK7g/edit)\n",
        "\n",
        "We also encourage users to join the [Disco Diffusion User Discord](https://discord.gg/XGZrFFCRfN) to learn from the active user community.\n",
        "\n",
        "This section below is outdated as of v2\n",
        "\n",
        "Setting | Description | Default\n",
        "--- | --- | ---\n",
        "**Your vision:**\n",
        "`text_prompts` | A description of what you'd like the machine to generate. Think of it like writing the caption below your image on a website. | `N/A`\n",
        "`image_prompts` | Think of these images more as a description of their contents. | `N/A`\n",
        "**Image quality:**\n",
        "`clip_guidance_scale`  | Controls how much the image should look like the prompt. | `1000`\n",
        "`tv_scale` | Controls the smoothness of the final output. | `150`\n",
        "`range_scale` | Controls how far out of range RGB values are allowed to be. | `150`\n",
        "`sat_scale` | Controls how much saturation is allowed. From nshepperd's JAX notebook. | `0`\n",
        "`cutn` | Controls how many crops to take from the image. | `16`\n",
        "`cutn_batches` | Accumulate CLIP gradient from multiple batches of cuts. | `2`\n",
        "**Go BIG Settings:**\n",
        "`go_big` | A boolean to enable Go BIG tiled upscale diffusion. | `False`\n",
        "`go_big_scale` | The Go BIG multiplier factor | `2`+\n",
        "`gobig_skip_perent` | Percentage of steps to skip when doing Go BIG tiled diffusion runs. | `85`\n",
        "`go_big_init` | A path or URL to a Go BIG init image. This will skip the diffusion run. | `None`\n",
        "`go_big_init_skipresize` | Whether to skip resizing the go BIG init image | `False`\n",
        "`render_mask` | A alpha mask of areas to diffuse (white), and not diffuse (black) | `None`\n",
        "**Init settings:**\n",
        "`init_image` | URL or local path | `None`\n",
        "`init_scale` | This enhances the effect of the init image, a good value is 1000 | `0`\n",
        "`skip_steps` | Controls the starting point along the diffusion timesteps | `0`\n",
        "`perlin_init` | Option to start with random perlin noise | `False`\n",
        "`perlin_mode` | (`gray`, `color`) | `mixed`\n",
        "**Advanced:**\n",
        "`skip_augs` | Controls whether to skip torchvision augmentations | `False`\n",
        "`randomize_class` | Controls whether the imagenet class is randomly changed each iteration | `True`\n",
        "`clip_denoised` | Determines whether CLIP discriminates a noisy or denoised image | `False`\n",
        "`clamp_grad` | Experimental: Using adaptive clip grad in the cond_fn | `True`\n",
        "`seed`  | Choose a random seed and print it at end of run for reproduction | `random_seed`\n",
        "`fuzzy_prompt` | Controls whether to add multiple noisy prompts to the prompt losses | `False`\n",
        "`rand_mag` | Controls the magnitude of the random noise | `0.1`\n",
        "`eta` | DDIM hyperparameter | `0.5`\n",
        "`use_vertical_symmetry` | Enforce symmetry over x axis of the image on [`tr_st`*`steps` for `tr_st` in `transformation_steps`] steps of the diffusion process | `False`\n",
        "`use_horizontal_symmetry` | Enforce symmetry over y axis of the image on [`tr_st`*`steps` for `tr_st` in `transformation_steps`] steps of the diffusion process | `False`\n",
        "`transformation_steps` | Steps (expressed in percentages) in which the symmetry is enforced | `[0.01]`\n",
        "`video_init_flow_warp` | Flow warp enabled | True\n",
        "`video_init_flow_blend` | 0 - you get raw input, 1 - you get warped diffused previous frame  | `0.999`\n",
        "`video_init_check_consistency` | TBD check forward-backward flow consistency (uncheck unless there are too many warping artifacts) | `False`\n",
        "\n",
        "<br>\n",
        "\n",
        "<font size=\"5\">**Model settings**</font>\n",
        "<hr>\n",
        "\n",
        "Setting | Description | Default\n",
        "--- | --- | ---\n",
        "**Diffusion:**\n",
        "`timestep_respacing` | Modify this value to decrease the number of timesteps. | ddim100\n",
        "`diffusion_steps` || 1000\n",
        "**Diffusion:**\n",
        "`clip_models` | Models of CLIP to load. Typically the more, the better but they all come at a hefty VRAM cost. | ViT-B/32, ViT-B/16, RN50x4\n",
        "\n",
        "## Noodle Soup Prompts\n",
        "\n",
        "To use a term database, simply use one of keys below. \n",
        "\n",
        "For example if you wanted beauty adjective, you would write `_adj-beauty_` in your prompt. \n",
        "\n",
        "## Terminology Keys\n",
        "\n",
        "### &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; Adjective Types\n",
        "   - `_adj-architecture_` - A list of architectural adjectives and styles\n",
        "   - `_adj-beauty_` - A list of beauty adjectives for people (maybe things?)\n",
        "   - `_adj-general_` - A list of general adjectives for people/things.\n",
        "   - `_adj-horror_` - A list of horror adjectives\n",
        "### Art Types\n",
        "   - `_artist_` - A comprehensive list of artists by **MisterRuffian** <font size=\"2\">(<font color=\"7e87f6\">Discord</font> Misterruffian#2891)</font>\n",
        "   - `_color_` - A comprehensive list of colors\n",
        "   - `_style_` - A list of art styles and mediums\n",
        "### Computer Graphics Types\n",
        "   - `_3d-terms_` - A list of 3D graphics terminology\n",
        "   - `_color-palette_` - A list of computer and video game console color palettes\n",
        "   - `_hd_` - A list of high definition resolution terms\n",
        "### Miscellaneous Types\n",
        "   - `_site_` - A list of websites to query\n",
        "   - `_nationality_` - A list of nationalities\n",
        "   - `_gen-modififer_` - A list of general modifiers adopted from [Weird Wonderful AI Art](https://weirdwonderfulai.art/)\n",
        "### Noun Types\n",
        "   - `_noun-beauty_` - A list of beauty related nouns\n",
        "   - `_noun-emote_` - A list of emotions and expressions\n",
        "   - `_noun-fantasy_` - A list of fantasy nouns\n",
        "   - `_noun-general_` - A list of general nouns\n",
        "   - `_noun-horror_` - A list of horror nouns\n",
        "### Photography / Image / Film Types\n",
        "   - `_aspect-ratio_` - A list of common aspect ratios\n",
        "   - `_cameras_` - A list of camera models *(including manufactuerer)*\n",
        "   - `_camera-manu_` - A list of camera manufacturers\n",
        "   - `_f-stop_` - A list of camera aperture f-stop\n",
        "   - `_focal-length_` - A list of focal length ranges\n",
        "   - `_photo-term_` - A list of photography terms relating to photos\n",
        "\n",
        "<br>\n",
        "\n",
        "## **Easing Function Aliases**\n",
        " - Easing Integer: `easeint(st=40, en=1, s=800, fn=cut_easing_fn)` \n",
        " - Easing Float: `easeflt(st=40, en=1, s=800, fn=cut_easing_fn)`\n",
        "<br><br>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SetupTop"
      },
      "source": [
        "# 1. Set Up"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "CheckGPU"
      },
      "source": [
        "#@title 1.1 Check GPU Status\n",
        "import subprocess\n",
        "simple_nvidia_smi_display = False#@param {type:\"boolean\"}\n",
        "nvidiasmi_simple = subprocess.run(['nvidia-smi', '-L'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "if simple_nvidia_smi_display:\n",
        "    #!nvidia-smi\n",
        "    print(nvidiasmi_simple)\n",
        "else:\n",
        "    #!nvidia-smi -i 0 -e 0\n",
        "    nvidiasmi_output = subprocess.run(['nvidia-smi'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(nvidiasmi_output)\n",
        "    nvidiasmi_ecc_note = subprocess.run(['nvidia-smi', '-i', '0', '-e', '0'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(nvidiasmi_ecc_note)\n",
        "\n",
        "gpu_name = nvidiasmi_simple.split(':')[1].split('(')[0].strip()\n",
        "\n",
        "gpu_warning = None\n",
        "if 'Tesla T4' or 'Tesla P100' in gpu_name:\n",
        "  gpu_warning = \"\\33[33m\\u001b[4mWARNING:\\33[0m \\033[1mYou are using a limited VRAM GPU (\\33[33m\"+gpu_name+\"\\33[0m\\033[1m). \\n         Be cautious when using resolutions above 720p, or heavier models such ViTL14+, RN50x4+, etc.\\33[0m\"\n",
        "  print(gpu_warning)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "PrepFolders"
      },
      "source": [
        "#@title 1.2 Setup DiscoStream Environment\n",
        "\n",
        "# PREPARE FOLDERS AND GDRIVE\n",
        "\n",
        "import subprocess, os, sys, ipykernel\n",
        "\n",
        "def gitclone(url, targetdir=None):\n",
        "    if targetdir:\n",
        "        res = subprocess.run(['git', 'clone', url, targetdir], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    else:\n",
        "        res = subprocess.run(['git', 'clone', url], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(res)\n",
        "\n",
        "def pipi(modulestr):\n",
        "    res = subprocess.run(['pip', 'install', modulestr], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(res)\n",
        "\n",
        "def pipie(modulestr):\n",
        "    res = subprocess.run(['git', 'install', '-e', modulestr], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(res)\n",
        "\n",
        "def wget(url, outputdir):\n",
        "    res = subprocess.run(['wget', url, '-P', f'{outputdir}'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(res)\n",
        "\n",
        "try:\n",
        "    from google.colab import drive\n",
        "    print(\"Google Colab detected. Using Google Drive.\")\n",
        "    is_colab = True\n",
        "    #@markdown If you connect your Google Drive, you can save the final image of each run on your drive.\n",
        "    google_drive = True #@param {type:\"boolean\"}\n",
        "    #@markdown Click here if you'd like to save the diffusion model checkpoint file to (and/or load from) your Google Drive:\n",
        "    save_models_to_google_drive = True #@param {type:\"boolean\"}\n",
        "except:\n",
        "    is_colab = False\n",
        "    google_drive = False\n",
        "    save_models_to_google_drive = False\n",
        "    print(\"Google Colab not detected.\")\n",
        "\n",
        "if is_colab:\n",
        "    if google_drive is True:\n",
        "        drive.mount('/content/drive')\n",
        "        root_path = '/content/drive/MyDrive/AI/Disco_Diffusion'\n",
        "    else:\n",
        "        root_path = '/content'\n",
        "else:\n",
        "    root_path = os.getcwd()\n",
        "\n",
        "import os\n",
        "def createPath(filepath):\n",
        "    os.makedirs(filepath, exist_ok=True)\n",
        "\n",
        "initDirPath = f'{root_path}/init_images'\n",
        "createPath(initDirPath)\n",
        "outDirPath = f'{root_path}/images_out'\n",
        "createPath(outDirPath)\n",
        "\n",
        "if is_colab:\n",
        "    if google_drive and not save_models_to_google_drive or not google_drive:\n",
        "        model_path = '/content/models'\n",
        "        createPath(model_path)\n",
        "    if google_drive and save_models_to_google_drive:\n",
        "        model_path = f'{root_path}/models'\n",
        "        createPath(model_path)\n",
        "else:\n",
        "    model_path = f'{root_path}/models'\n",
        "    createPath(model_path)\n",
        "\n",
        "# libraries = f'{root_path}/libraries'\n",
        "# createPath(libraries)\n",
        "\n",
        "# INSTALL DEPENDENCIES\n",
        "\n",
        "import pathlib, shutil, os, sys\n",
        "\n",
        "# There are some reports that with a T4 or V100 on Colab, downgrading to a previous version of PyTorch may be necessary.\n",
        "# .. but there are also reports that downgrading breaks them!  If you're facing issues, you may want to try uncommenting and running this code.\n",
        "# nvidiasmi_output = subprocess.run(['nvidia-smi'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "# cards_requiring_downgrade = [\"Tesla T4\", \"V100\"]\n",
        "# if is_colab:\n",
        "#     if any(cardstr in nvidiasmi_output for cardstr in cards_requiring_downgrade):\n",
        "#         print(\"Downgrading pytorch. This can take a couple minutes ...\")\n",
        "#         downgrade_pytorch_result = subprocess.run(['pip', 'install', 'torch==1.10.2', 'torchvision==0.11.3', '-q'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "#         print(\"pytorch downgraded.\")\n",
        "\n",
        "#@markdown Check this if you want to use CPU\n",
        "useCPU = False #@param {type:\"boolean\"}\n",
        "\n",
        "if not is_colab:\n",
        "    # If running locally, there's a good chance your env will need this in order to not crash upon np.matmul() or similar operations.\n",
        "    os.environ['KMP_DUPLICATE_LIB_OK']='TRUE'\n",
        "\n",
        "PROJECT_DIR = os.path.abspath(os.getcwd())\n",
        "USE_ADABINS = True\n",
        "\n",
        "if is_colab:\n",
        "    if not google_drive:\n",
        "        root_path = f'/content'\n",
        "        model_path = '/content/models' \n",
        "else:\n",
        "    root_path = os.getcwd()\n",
        "    model_path = f'{root_path}/models'\n",
        "\n",
        "multipip_res = subprocess.run(['pip', 'install', 'lpips', 'datetime', 'timm', 'ftfy', 'einops', 'pytorch-lightning', 'omegaconf'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "print(multipip_res)\n",
        "\n",
        "if is_colab:\n",
        "    subprocess.run(['apt', 'install', 'imagemagick'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "\n",
        "try:\n",
        "    from CLIP import clip\n",
        "except:\n",
        "    if not os.path.exists(\"CLIP\"):\n",
        "        gitclone(\"https://github.com/openai/CLIP\")\n",
        "    sys.path.append(f'{PROJECT_DIR}/CLIP')\n",
        "\n",
        "try:\n",
        "    import open_clip\n",
        "except:\n",
        "    if not os.path.exists(\"open_clip/src\"):\n",
        "        gitclone(\"https://github.com/mlfoundations/open_clip.git\")\n",
        "    sys.path.append(f'{PROJECT_DIR}/open_clip/src')\n",
        "    import open_clip\n",
        "\n",
        "try:\n",
        "    from guided_diffusion.script_util import create_model_and_diffusion\n",
        "except:\n",
        "    if not os.path.exists(\"guided-diffusion\"):\n",
        "        gitclone(\"https://github.com/kostarion/guided-diffusion\")\n",
        "    sys.path.append(f'{PROJECT_DIR}/guided-diffusion')\n",
        "\n",
        "try:\n",
        "    from resize_right import resize\n",
        "except:\n",
        "    if not os.path.exists(\"ResizeRight\"):\n",
        "        gitclone(\"https://github.com/assafshocher/ResizeRight.git\")\n",
        "    sys.path.append(f'{PROJECT_DIR}/ResizeRight')\n",
        "\n",
        "try:\n",
        "    import py3d_tools\n",
        "except:\n",
        "    if not os.path.exists('pytorch3d-lite'):\n",
        "        gitclone(\"https://github.com/MSFTserver/pytorch3d-lite.git\")\n",
        "    sys.path.append(f'{PROJECT_DIR}/pytorch3d-lite')\n",
        "\n",
        "try:\n",
        "    from midas.dpt_depth import DPTDepthModel\n",
        "except:\n",
        "    if not os.path.exists('MiDaS'):\n",
        "        gitclone(\"https://github.com/isl-org/MiDaS.git\")\n",
        "    if not os.path.exists('MiDaS/midas_utils.py'):\n",
        "        shutil.move('MiDaS/utils.py', 'MiDaS/midas_utils.py')\n",
        "    if not os.path.exists(f'{model_path}/dpt_large-midas-2f21e586.pt'):\n",
        "        wget(\"https://github.com/intel-isl/DPT/releases/download/1_0/dpt_large-midas-2f21e586.pt\", model_path)\n",
        "    sys.path.append(f'{PROJECT_DIR}/MiDaS')\n",
        "\n",
        "try:\n",
        "    sys.path.append(PROJECT_DIR)\n",
        "    import disco_xform_utils as dxf\n",
        "except:\n",
        "    if not os.path.exists(\"disco-diffusion\"):\n",
        "        gitclone(\"https://github.com/alembics/disco-diffusion.git\")\n",
        "    if not os.path.exists('disco_xform_utils.py'):\n",
        "        shutil.move('disco-diffusion/disco_xform_utils.py', 'disco_xform_utils.py')\n",
        "    sys.path.append(PROJECT_DIR)\n",
        "\n",
        "import torch\n",
        "from dataclasses import dataclass\n",
        "from functools import partial\n",
        "import cv2\n",
        "import pandas as pd\n",
        "import gc\n",
        "import io\n",
        "import math\n",
        "import timm\n",
        "from IPython import display\n",
        "import lpips\n",
        "from PIL import Image, ImageOps\n",
        "import requests\n",
        "from glob import glob\n",
        "import json\n",
        "from types import SimpleNamespace\n",
        "from torch import nn\n",
        "from torch.nn import functional as F\n",
        "import torchvision.transforms as T\n",
        "import torchvision.transforms.functional as TF\n",
        "from tqdm.notebook import tqdm\n",
        "from CLIP import clip\n",
        "from resize_right import resize\n",
        "from guided_diffusion.script_util import create_model_and_diffusion, model_and_diffusion_defaults\n",
        "from datetime import datetime\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "import random\n",
        "from ipywidgets import Output\n",
        "import hashlib\n",
        "from functools import partial\n",
        "if is_colab:\n",
        "    os.chdir('/content')\n",
        "    from google.colab import files\n",
        "else:\n",
        "    os.chdir(f'{PROJECT_DIR}')\n",
        "from IPython.display import Image as ipyimg\n",
        "from numpy import asarray\n",
        "from einops import rearrange, repeat\n",
        "import torch, torchvision\n",
        "import time\n",
        "from omegaconf import OmegaConf\n",
        "import warnings\n",
        "warnings.filterwarnings(\"ignore\", category=UserWarning)\n",
        "import json\n",
        "from os.path import exists\n",
        "import pprint\n",
        "from urllib.parse import urlparse\n",
        "\n",
        "# Noodle Soup prompts\n",
        "!wget -q --show-progress --no-cache --backups=1 'https://raw.githubusercontent.com/WASasquatch/noodle-soup-prompts/main/nsp_pantry.py'\n",
        "import nsp_pantry\n",
        "from nsp_pantry import nspterminology, nsp_parse\n",
        "nsp = False\n",
        "use_this_prompt = False\n",
        "\n",
        "if nsp_parse and nspterminology:\n",
        "  print(\"\\33[32mNoodle Soup Prompts installed successfuly.\\33[0m \\x1B[3mMmm. Noodle Soup.\\x1B[0m\")\n",
        "  nsp = True\n",
        "\n",
        "# AdaBins stuff\n",
        "if USE_ADABINS:\n",
        "    try:\n",
        "        from infer import InferenceHelper\n",
        "    except:\n",
        "        if not os.path.exists(\"AdaBins\"):\n",
        "            gitclone(\"https://github.com/shariqfarooq123/AdaBins.git\")\n",
        "        if not os.path.exists(f'{PROJECT_DIR}/pretrained/AdaBins_nyu.pt'):\n",
        "            createPath(f'{PROJECT_DIR}/pretrained')\n",
        "            wget(\"https://cloudflare-ipfs.com/ipfs/Qmd2mMnDLWePKmgfS8m6ntAg4nhV5VkUyAydYBp8cWWeB7/AdaBins_nyu.pt\", f'{PROJECT_DIR}/pretrained')\n",
        "        sys.path.append(f'{PROJECT_DIR}/AdaBins')\n",
        "    from infer import InferenceHelper\n",
        "    MAX_ADABINS_AREA = 500000\n",
        "\n",
        "import torch\n",
        "DEVICE = torch.device('cuda:0' if (torch.cuda.is_available() and not useCPU) else 'cpu')\n",
        "print('Using device:', DEVICE)\n",
        "device = DEVICE # At least one of the modules expects this name..\n",
        "\n",
        "if not useCPU:\n",
        "    if torch.cuda.get_device_capability(DEVICE) == (8,0): ## A100 fix thanks to Emad\n",
        "        print('Disabling CUDNN for A100 gpu', file=sys.stderr)\n",
        "        torch.backends.cudnn.enabled = False\n",
        "\n",
        "# DEFINE MIDAS FUNCTION\n",
        "\n",
        "from midas.dpt_depth import DPTDepthModel\n",
        "from midas.midas_net import MidasNet\n",
        "from midas.midas_net_custom import MidasNet_small\n",
        "from midas.transforms import Resize, NormalizeImage, PrepareForNet\n",
        "\n",
        "# Initialize MiDaS depth model.\n",
        "# It remains resident in VRAM and likely takes around 2GB VRAM.\n",
        "# You could instead initialize it for each frame (and free it after each frame) to save VRAM.. but initializing it is slow.\n",
        "default_models = {\n",
        "    \"midas_v21_small\": f\"{model_path}/midas_v21_small-70d6b9c8.pt\",\n",
        "    \"midas_v21\": f\"{model_path}/midas_v21-f6b98070.pt\",\n",
        "    \"dpt_large\": f\"{model_path}/dpt_large-midas-2f21e586.pt\",\n",
        "    \"dpt_hybrid\": f\"{model_path}/dpt_hybrid-midas-501f0c75.pt\",\n",
        "    \"dpt_hybrid_nyu\": f\"{model_path}/dpt_hybrid_nyu-2ce69ec7.pt\",}\n",
        "\n",
        "\n",
        "def init_midas_depth_model(midas_model_type=\"dpt_large\", optimize=True):\n",
        "    midas_model = None\n",
        "    net_w = None\n",
        "    net_h = None\n",
        "    resize_mode = None\n",
        "    normalization = None\n",
        "\n",
        "    print(f\"Initializing MiDaS '{midas_model_type}' depth model...\")\n",
        "    # load network\n",
        "    midas_model_path = default_models[midas_model_type]\n",
        "\n",
        "    if midas_model_type == \"dpt_large\": # DPT-Large\n",
        "        midas_model = DPTDepthModel(\n",
        "            path=midas_model_path,\n",
        "            backbone=\"vitl16_384\",\n",
        "            non_negative=True,\n",
        "        )\n",
        "        net_w, net_h = 384, 384\n",
        "        resize_mode = \"minimal\"\n",
        "        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])\n",
        "    elif midas_model_type == \"dpt_hybrid\": #DPT-Hybrid\n",
        "        midas_model = DPTDepthModel(\n",
        "            path=midas_model_path,\n",
        "            backbone=\"vitb_rn50_384\",\n",
        "            non_negative=True,\n",
        "        )\n",
        "        net_w, net_h = 384, 384\n",
        "        resize_mode=\"minimal\"\n",
        "        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])\n",
        "    elif midas_model_type == \"dpt_hybrid_nyu\": #DPT-Hybrid-NYU\n",
        "        midas_model = DPTDepthModel(\n",
        "            path=midas_model_path,\n",
        "            backbone=\"vitb_rn50_384\",\n",
        "            non_negative=True,\n",
        "        )\n",
        "        net_w, net_h = 384, 384\n",
        "        resize_mode=\"minimal\"\n",
        "        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])\n",
        "    elif midas_model_type == \"midas_v21\":\n",
        "        midas_model = MidasNet(midas_model_path, non_negative=True)\n",
        "        net_w, net_h = 384, 384\n",
        "        resize_mode=\"upper_bound\"\n",
        "        normalization = NormalizeImage(\n",
        "            mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]\n",
        "        )\n",
        "    elif midas_model_type == \"midas_v21_small\":\n",
        "        midas_model = MidasNet_small(midas_model_path, features=64, backbone=\"efficientnet_lite3\", exportable=True, non_negative=True, blocks={'expand': True})\n",
        "        net_w, net_h = 256, 256\n",
        "        resize_mode=\"upper_bound\"\n",
        "        normalization = NormalizeImage(\n",
        "            mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]\n",
        "        )\n",
        "    else:\n",
        "        print(f\"midas_model_type '{midas_model_type}' not implemented\")\n",
        "        assert False\n",
        "\n",
        "    midas_transform = T.Compose(\n",
        "        [\n",
        "            Resize(\n",
        "                net_w,\n",
        "                net_h,\n",
        "                resize_target=None,\n",
        "                keep_aspect_ratio=True,\n",
        "                ensure_multiple_of=32,\n",
        "                resize_method=resize_mode,\n",
        "                image_interpolation_method=cv2.INTER_CUBIC,\n",
        "            ),\n",
        "            normalization,\n",
        "            PrepareForNet(),\n",
        "        ]\n",
        "    )\n",
        "\n",
        "    midas_model.eval()\n",
        "    \n",
        "    if optimize==True:\n",
        "        if DEVICE == torch.device(\"cuda\"):\n",
        "            midas_model = midas_model.to(memory_format=torch.channels_last)  \n",
        "            midas_model = midas_model.half()\n",
        "\n",
        "    midas_model.to(DEVICE)\n",
        "\n",
        "    print(f\"MiDaS '{midas_model_type}' depth model initialized.\")\n",
        "    return midas_model, midas_transform, net_w, net_h, resize_mode, normalization\n",
        "\n",
        "# DEFINE MAIN FUNCTIONS\n",
        "\n",
        "# https://gist.github.com/adefossez/0646dbe9ed4005480a2407c62aac8869\n",
        "\n",
        "import py3d_tools as p3dT\n",
        "import disco_xform_utils as dxf\n",
        "\n",
        "def interp(t):\n",
        "    return 3 * t**2 - 2 * t ** 3\n",
        "\n",
        "def perlin(width, height, scale=10, device=None):\n",
        "    gx, gy = torch.randn(2, width + 1, height + 1, 1, 1, device=device)\n",
        "    xs = torch.linspace(0, 1, scale + 1)[:-1, None].to(device)\n",
        "    ys = torch.linspace(0, 1, scale + 1)[None, :-1].to(device)\n",
        "    wx = 1 - interp(xs)\n",
        "    wy = 1 - interp(ys)\n",
        "    dots = 0\n",
        "    dots += wx * wy * (gx[:-1, :-1] * xs + gy[:-1, :-1] * ys)\n",
        "    dots += (1 - wx) * wy * (-gx[1:, :-1] * (1 - xs) + gy[1:, :-1] * ys)\n",
        "    dots += wx * (1 - wy) * (gx[:-1, 1:] * xs - gy[:-1, 1:] * (1 - ys))\n",
        "    dots += (1 - wx) * (1 - wy) * (-gx[1:, 1:] * (1 - xs) - gy[1:, 1:] * (1 - ys))\n",
        "    return dots.permute(0, 2, 1, 3).contiguous().view(width * scale, height * scale)\n",
        "\n",
        "def perlin_ms(octaves, width, height, grayscale, device=device):\n",
        "    out_array = [0.5] if grayscale else [0.5, 0.5, 0.5]\n",
        "    # out_array = [0.0] if grayscale else [0.0, 0.0, 0.0]\n",
        "    for i in range(1 if grayscale else 3):\n",
        "        scale = 2 ** len(octaves)\n",
        "        oct_width = width\n",
        "        oct_height = height\n",
        "        for oct in octaves:\n",
        "            p = perlin(oct_width, oct_height, scale, device)\n",
        "            out_array[i] += p * oct\n",
        "            scale //= 2\n",
        "            oct_width *= 2\n",
        "            oct_height *= 2\n",
        "    return torch.cat(out_array)\n",
        "\n",
        "def create_perlin_noise(octaves=[1, 1, 1, 1], width=2, height=2, grayscale=True):\n",
        "    out = perlin_ms(octaves, width, height, grayscale)\n",
        "    if grayscale:\n",
        "        out = TF.resize(size=(side_y, side_x), img=out.unsqueeze(0))\n",
        "        out = TF.to_pil_image(out.clamp(0, 1)).convert('RGB')\n",
        "    else:\n",
        "        out = out.reshape(-1, 3, out.shape[0]//3, out.shape[1])\n",
        "        out = TF.resize(size=(side_y, side_x), img=out)\n",
        "        out = TF.to_pil_image(out.clamp(0, 1).squeeze())\n",
        "\n",
        "    out = ImageOps.autocontrast(out)\n",
        "    return out\n",
        "\n",
        "def regen_perlin():\n",
        "    if perlin_mode == 'color':\n",
        "        init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, False)\n",
        "        init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, False)\n",
        "    elif perlin_mode == 'gray':\n",
        "        init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, True)\n",
        "        init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, True)\n",
        "    else:\n",
        "        init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, False)\n",
        "        init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, True)\n",
        "\n",
        "    init = TF.to_tensor(init).add(TF.to_tensor(init2)).div(2).to(device).unsqueeze(0).mul(2).sub(1)\n",
        "    del init2\n",
        "    return init.expand(batch_size, -1, -1, -1)\n",
        "\n",
        "def fetch(url_or_path):\n",
        "    if str(url_or_path).startswith('http://') or str(url_or_path).startswith('https://'):\n",
        "        r = requests.get(url_or_path)\n",
        "        r.raise_for_status()\n",
        "        fd = io.BytesIO()\n",
        "        fd.write(r.content)\n",
        "        fd.seek(0)\n",
        "        return fd\n",
        "    return open(url_or_path, 'rb')\n",
        "\n",
        "def read_image_workaround(path):\n",
        "    \"\"\"OpenCV reads images as BGR, Pillow saves them as RGB. Work around\n",
        "    this incompatibility to avoid colour inversions.\"\"\"\n",
        "    im_tmp = cv2.imread(path)\n",
        "    return cv2.cvtColor(im_tmp, cv2.COLOR_BGR2RGB)\n",
        "\n",
        "def parse_prompt(prompt):\n",
        "    if prompt.startswith('http://') or prompt.startswith('https://'):\n",
        "        vals = prompt.rsplit(':', 2)\n",
        "        vals = [vals[0] + ':' + vals[1], *vals[2:]]\n",
        "    else:\n",
        "        vals = prompt.rsplit(':', 1)\n",
        "    vals = vals + ['', '1'][len(vals):]\n",
        "    vals[1] = eval(vals[1])\n",
        "    if isinstance(vals[1], int) or isinstance(vals[1], float):\n",
        "      return vals[0], [ float(x) for x in [float(vals[1])]*1000 ]\n",
        "    elif isinstance(vals[1], list):\n",
        "      return vals[0], [ float(x) for x in vals[1] ]\n",
        "    else:\n",
        "      raise TypeError(\"Only strings that evaluate to integers, floats and lists are allowed\")\n",
        "\n",
        "def sinc(x):\n",
        "    return torch.where(x != 0, torch.sin(math.pi * x) / (math.pi * x), x.new_ones([]))\n",
        "\n",
        "def lanczos(x, a):\n",
        "    cond = torch.logical_and(-a < x, x < a)\n",
        "    out = torch.where(cond, sinc(x) * sinc(x/a), x.new_zeros([]))\n",
        "    return out / out.sum()\n",
        "\n",
        "def ramp(ratio, width):\n",
        "    n = math.ceil(width / ratio + 1)\n",
        "    out = torch.empty([n])\n",
        "    cur = 0\n",
        "    for i in range(out.shape[0]):\n",
        "        out[i] = cur\n",
        "        cur += ratio\n",
        "    return torch.cat([-out[1:].flip([0]), out])[1:-1]\n",
        "\n",
        "def resample(input, size, align_corners=True):\n",
        "    n, c, h, w = input.shape\n",
        "    dh, dw = size\n",
        "\n",
        "    input = input.reshape([n * c, 1, h, w])\n",
        "\n",
        "    if dh < h:\n",
        "        kernel_h = lanczos(ramp(dh / h, 2), 2).to(input.device, input.dtype)\n",
        "        pad_h = (kernel_h.shape[0] - 1) // 2\n",
        "        input = F.pad(input, (0, 0, pad_h, pad_h), 'reflect')\n",
        "        input = F.conv2d(input, kernel_h[None, None, :, None])\n",
        "\n",
        "    if dw < w:\n",
        "        kernel_w = lanczos(ramp(dw / w, 2), 2).to(input.device, input.dtype)\n",
        "        pad_w = (kernel_w.shape[0] - 1) // 2\n",
        "        input = F.pad(input, (pad_w, pad_w, 0, 0), 'reflect')\n",
        "        input = F.conv2d(input, kernel_w[None, None, None, :])\n",
        "\n",
        "    input = input.reshape([n, c, h, w])\n",
        "    return F.interpolate(input, size, mode='bicubic', align_corners=align_corners)\n",
        "\n",
        "class MakeCutouts(nn.Module):\n",
        "    def __init__(self, cut_size, cutn, skip_augs=False):\n",
        "        super().__init__()\n",
        "        self.cut_size = cut_size\n",
        "        self.cutn = cutn\n",
        "        self.skip_augs = skip_augs\n",
        "        self.augs = T.Compose([\n",
        "            T.RandomHorizontalFlip(p=0.5),\n",
        "            T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "            T.RandomAffine(degrees=15, translate=(0.1, 0.1)),\n",
        "            T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "            T.RandomPerspective(distortion_scale=0.4, p=0.7),\n",
        "            T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "            T.RandomGrayscale(p=0.15),\n",
        "            T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "            # T.ColorJitter(brightness=0.1, contrast=0.1, saturation=0.1, hue=0.1),\n",
        "        ])\n",
        "\n",
        "    def forward(self, input):\n",
        "        input = T.Pad(input.shape[2]//4, fill=0)(input)\n",
        "        sideY, sideX = input.shape[2:4]\n",
        "        max_size = min(sideX, sideY)\n",
        "\n",
        "        cutouts = []\n",
        "        for ch in range(self.cutn):\n",
        "            if ch > self.cutn - self.cutn//4:\n",
        "                cutout = input.clone()\n",
        "            else:\n",
        "                size = int(max_size * torch.zeros(1,).normal_(mean=.8, std=.3).clip(float(self.cut_size/max_size), 1.))\n",
        "                offsetx = torch.randint(0, abs(sideX - size + 1), ())\n",
        "                offsety = torch.randint(0, abs(sideY - size + 1), ())\n",
        "                cutout = input[:, :, offsety:offsety + size, offsetx:offsetx + size]\n",
        "\n",
        "            if not self.skip_augs:\n",
        "                cutout = self.augs(cutout)\n",
        "            cutouts.append(resample(cutout, (self.cut_size, self.cut_size)))\n",
        "            del cutout\n",
        "\n",
        "        cutouts = torch.cat(cutouts, dim=0)\n",
        "        return cutouts\n",
        "\n",
        "cutout_debug = False\n",
        "padargs = {}\n",
        "\n",
        "class MakeCutoutsDango(nn.Module):\n",
        "    def __init__(self, cut_size,\n",
        "                 Overview=4, \n",
        "                 InnerCrop = 0, IC_Size_Pow=0.5, IC_Grey_P = 0.2\n",
        "                 ):\n",
        "        super().__init__()\n",
        "        self.cut_size = cut_size\n",
        "        self.Overview = Overview\n",
        "        self.InnerCrop = InnerCrop\n",
        "        self.IC_Size_Pow = IC_Size_Pow\n",
        "        self.IC_Grey_P = IC_Grey_P\n",
        "        if args.animation_mode == 'None':\n",
        "          self.augs = T.Compose([\n",
        "              T.RandomHorizontalFlip(p=0.5),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomAffine(degrees=10, translate=(0.05, 0.05),  interpolation = T.InterpolationMode.BILINEAR),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomGrayscale(p=0.1),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.ColorJitter(brightness=0.1, contrast=0.1, saturation=0.1, hue=0.1),\n",
        "          ])\n",
        "        elif args.animation_mode == 'Video Input':\n",
        "          self.augs = T.Compose([\n",
        "              T.RandomHorizontalFlip(p=0.5),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomAffine(degrees=15, translate=(0.1, 0.1)),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomPerspective(distortion_scale=0.4, p=0.7),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomGrayscale(p=0.15),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              # T.ColorJitter(brightness=0.1, contrast=0.1, saturation=0.1, hue=0.1),\n",
        "          ])\n",
        "        elif  args.animation_mode == '2D' or args.animation_mode == '3D':\n",
        "          self.augs = T.Compose([\n",
        "              T.RandomHorizontalFlip(p=0.4),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomAffine(degrees=10, translate=(0.05, 0.05),  interpolation = T.InterpolationMode.BILINEAR),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomGrayscale(p=0.1),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.ColorJitter(brightness=0.1, contrast=0.1, saturation=0.1, hue=0.3),\n",
        "          ])\n",
        "          \n",
        "\n",
        "    def forward(self, input):\n",
        "        cutouts = []\n",
        "        gray = T.Grayscale(3)\n",
        "        sideY, sideX = input.shape[2:4]\n",
        "        max_size = min(sideX, sideY)\n",
        "        min_size = min(sideX, sideY, self.cut_size)\n",
        "        l_size = max(sideX, sideY)\n",
        "        output_shape = [1,3,self.cut_size,self.cut_size] \n",
        "        output_shape_2 = [1,3,self.cut_size+2,self.cut_size+2]\n",
        "        pad_input = F.pad(input,((sideY-max_size)//2,(sideY-max_size)//2,(sideX-max_size)//2,(sideX-max_size)//2), **padargs)\n",
        "        cutout = resize(pad_input, out_shape=output_shape)\n",
        "\n",
        "        if self.Overview>0:\n",
        "            if self.Overview<=4:\n",
        "                if self.Overview>=1:\n",
        "                    cutouts.append(cutout)\n",
        "                if self.Overview>=2:\n",
        "                    cutouts.append(gray(cutout))\n",
        "                if self.Overview>=3:\n",
        "                    cutouts.append(TF.hflip(cutout))\n",
        "                if self.Overview==4:\n",
        "                    cutouts.append(gray(TF.hflip(cutout)))\n",
        "            else:\n",
        "                cutout = resize(pad_input, out_shape=output_shape)\n",
        "                for _ in range(self.Overview):\n",
        "                    cutouts.append(cutout)\n",
        "\n",
        "            if cutout_debug:\n",
        "                if is_colab:\n",
        "                    TF.to_pil_image(cutouts[0].clamp(0, 1).squeeze(0)).save(\"/content/cutout_overview0.jpg\",quality=99)\n",
        "                else:\n",
        "                    TF.to_pil_image(cutouts[0].clamp(0, 1).squeeze(0)).save(\"cutout_overview0.jpg\",quality=99)\n",
        "\n",
        "                              \n",
        "        if self.InnerCrop >0:\n",
        "            for i in range(self.InnerCrop):\n",
        "                size = int(torch.rand([])**self.IC_Size_Pow * (max_size - min_size) + min_size)\n",
        "                offsetx = torch.randint(0, sideX - size + 1, ())\n",
        "                offsety = torch.randint(0, sideY - size + 1, ())\n",
        "                cutout = input[:, :, offsety:offsety + size, offsetx:offsetx + size]\n",
        "                if i <= int(self.IC_Grey_P * self.InnerCrop):\n",
        "                    cutout = gray(cutout)\n",
        "                cutout = resize(cutout, out_shape=output_shape)\n",
        "                cutouts.append(cutout)\n",
        "            if cutout_debug:\n",
        "                if is_colab:\n",
        "                    TF.to_pil_image(cutouts[-1].clamp(0, 1).squeeze(0)).save(\"/content/cutout_InnerCrop.jpg\",quality=99)\n",
        "                else:\n",
        "                    TF.to_pil_image(cutouts[-1].clamp(0, 1).squeeze(0)).save(\"cutout_InnerCrop.jpg\",quality=99)\n",
        "        cutouts = torch.cat(cutouts)\n",
        "        if skip_augs is not True: cutouts=self.augs(cutouts)\n",
        "        return cutouts\n",
        "\n",
        "def spherical_dist_loss(x, y):\n",
        "    x = F.normalize(x, dim=-1)\n",
        "    y = F.normalize(y, dim=-1)\n",
        "    return (x - y).norm(dim=-1).div(2).arcsin().pow(2).mul(2)     \n",
        "\n",
        "def tv_loss(input):\n",
        "    \"\"\"L2 total variation loss, as in Mahendran et al.\"\"\"\n",
        "    input = F.pad(input, (0, 1, 0, 1), 'replicate')\n",
        "    x_diff = input[..., :-1, 1:] - input[..., :-1, :-1]\n",
        "    y_diff = input[..., 1:, :-1] - input[..., :-1, :-1]\n",
        "    return (x_diff**2 + y_diff**2).mean([1, 2, 3])\n",
        "\n",
        "\n",
        "def range_loss(input):\n",
        "    return (input - input.clamp(-1, 1)).pow(2).mean([1, 2, 3])\n",
        "\n",
        "stop_on_next_loop = False  # Make sure GPU memory doesn't get corrupted from cancelling the run mid-way through, allow a full frame to complete\n",
        "TRANSLATION_SCALE = 1.0/200.0\n",
        "\n",
        "def do_3d_step(img_filepath, frame_num, midas_model, midas_transform):\n",
        "  if args.key_frames:\n",
        "    translation_x = args.translation_x_series[frame_num]\n",
        "    translation_y = args.translation_y_series[frame_num]\n",
        "    translation_z = args.translation_z_series[frame_num]\n",
        "    rotation_3d_x = args.rotation_3d_x_series[frame_num]\n",
        "    rotation_3d_y = args.rotation_3d_y_series[frame_num]\n",
        "    rotation_3d_z = args.rotation_3d_z_series[frame_num]\n",
        "    print(\n",
        "        f'translation_x: {translation_x}',\n",
        "        f'translation_y: {translation_y}',\n",
        "        f'translation_z: {translation_z}',\n",
        "        f'rotation_3d_x: {rotation_3d_x}',\n",
        "        f'rotation_3d_y: {rotation_3d_y}',\n",
        "        f'rotation_3d_z: {rotation_3d_z}',\n",
        "    )\n",
        "\n",
        "  translate_xyz = [-translation_x*TRANSLATION_SCALE, translation_y*TRANSLATION_SCALE, -translation_z*TRANSLATION_SCALE]\n",
        "  rotate_xyz_degrees = [rotation_3d_x, rotation_3d_y, rotation_3d_z]\n",
        "  print('translation:',translate_xyz)\n",
        "  print('rotation:',rotate_xyz_degrees)\n",
        "  rotate_xyz = [math.radians(rotate_xyz_degrees[0]), math.radians(rotate_xyz_degrees[1]), math.radians(rotate_xyz_degrees[2])]\n",
        "  rot_mat = p3dT.euler_angles_to_matrix(torch.tensor(rotate_xyz, device=device), \"XYZ\").unsqueeze(0)\n",
        "  print(\"rot_mat: \" + str(rot_mat))\n",
        "  next_step_pil = dxf.transform_image_3d(img_filepath, midas_model, midas_transform, DEVICE,\n",
        "                                          rot_mat, translate_xyz, args.near_plane, args.far_plane,\n",
        "                                          args.fov, padding_mode=args.padding_mode,\n",
        "                                          sampling_mode=args.sampling_mode, midas_weight=args.midas_weight)\n",
        "  return next_step_pil\n",
        "\n",
        "def symmetry_transformation_fn(x):\n",
        "    if args.use_horizontal_symmetry:\n",
        "        [n, c, h, w] = x.size()\n",
        "        x = torch.concat((x[:, :, :, :w//2], torch.flip(x[:, :, :, :w//2], [-1])), -1)\n",
        "        print(\"horizontal symmetry applied\")\n",
        "    if args.use_vertical_symmetry:\n",
        "        [n, c, h, w] = x.size()\n",
        "        x = torch.concat((x[:, :, :h//2, :], torch.flip(x[:, :, :h//2, :], [-2])), -2)\n",
        "        print(\"vertical symmetry applied\")\n",
        "    return x\n",
        "\n",
        "def do_run():\n",
        "  seed = args.seed\n",
        "  print(range(args.start_frame, args.max_frames))\n",
        "\n",
        "  if (args.animation_mode == \"3D\") and (args.midas_weight > 0.0):\n",
        "      midas_model, midas_transform, midas_net_w, midas_net_h, midas_resize_mode, midas_normalization = init_midas_depth_model(args.midas_depth_model)\n",
        "  for frame_num in range(args.start_frame, args.max_frames):\n",
        "      if stop_on_next_loop:\n",
        "        break\n",
        "      \n",
        "      display.clear_output(wait=True)\n",
        "\n",
        "      # Print Frame progress if animation mode is on\n",
        "      if args.animation_mode != \"None\":\n",
        "        batchBar = tqdm(range(args.max_frames), desc =\"Frames\")\n",
        "        batchBar.n = frame_num\n",
        "        batchBar.refresh()\n",
        "\n",
        "      \n",
        "      # Inits if not video frames\n",
        "      if args.animation_mode != \"Video Input\":\n",
        "        if args.init_image in ['','none', 'None', 'NONE']:\n",
        "          init_image = None\n",
        "        else:\n",
        "          init_image = args.init_image\n",
        "        init_scale = args.init_scale\n",
        "        skip_steps = args.skip_steps\n",
        "\n",
        "      if args.animation_mode == \"2D\":\n",
        "        if args.key_frames:\n",
        "          angle = args.angle_series[frame_num]\n",
        "          zoom = args.zoom_series[frame_num]\n",
        "          translation_x = args.translation_x_series[frame_num]\n",
        "          translation_y = args.translation_y_series[frame_num]\n",
        "          print(\n",
        "              f'angle: {angle}',\n",
        "              f'zoom: {zoom}',\n",
        "              f'translation_x: {translation_x}',\n",
        "              f'translation_y: {translation_y}',\n",
        "          )\n",
        "        \n",
        "        if frame_num > 0:\n",
        "          seed += 1\n",
        "          if resume_run and frame_num == start_frame:\n",
        "            img_0 = cv2.imread(batchFolder+f\"/{batch_name}({batchNum})_{start_frame-1:04}.png\")\n",
        "          else:\n",
        "            img_0 = cv2.imread('prevFrame.png')\n",
        "          center = (1*img_0.shape[1]//2, 1*img_0.shape[0]//2)\n",
        "          trans_mat = np.float32(\n",
        "              [[1, 0, translation_x],\n",
        "              [0, 1, translation_y]]\n",
        "          )\n",
        "          rot_mat = cv2.getRotationMatrix2D( center, angle, zoom )\n",
        "          trans_mat = np.vstack([trans_mat, [0,0,1]])\n",
        "          rot_mat = np.vstack([rot_mat, [0,0,1]])\n",
        "          transformation_matrix = np.matmul(rot_mat, trans_mat)\n",
        "          img_0 = cv2.warpPerspective(\n",
        "              img_0,\n",
        "              transformation_matrix,\n",
        "              (img_0.shape[1], img_0.shape[0]),\n",
        "              borderMode=cv2.BORDER_WRAP\n",
        "          )\n",
        "\n",
        "          cv2.imwrite('prevFrameScaled.png', img_0)\n",
        "          init_image = 'prevFrameScaled.png'\n",
        "          init_scale = args.frames_scale\n",
        "          skip_steps = args.calc_frames_skip_steps\n",
        "\n",
        "      if args.animation_mode == \"3D\":\n",
        "        if frame_num > 0:\n",
        "          seed += 1    \n",
        "          if resume_run and frame_num == start_frame:\n",
        "            img_filepath = batchFolder+f\"/{batch_name}({batchNum})_{start_frame-1:04}.png\"\n",
        "            if turbo_mode and frame_num > turbo_preroll:\n",
        "              shutil.copyfile(img_filepath, 'oldFrameScaled.png')\n",
        "          else:\n",
        "            img_filepath = '/content/prevFrame.png' if is_colab else 'prevFrame.png'\n",
        "\n",
        "          next_step_pil = do_3d_step(img_filepath, frame_num, midas_model, midas_transform)\n",
        "          next_step_pil.save('prevFrameScaled.png')\n",
        "\n",
        "          ### Turbo mode - skip some diffusions, use 3d morph for clarity and to save time\n",
        "          if turbo_mode:\n",
        "            if frame_num == turbo_preroll: #start tracking oldframe\n",
        "              next_step_pil.save('oldFrameScaled.png')#stash for later blending          \n",
        "            elif frame_num > turbo_preroll:\n",
        "              #set up 2 warped image sequences, old & new, to blend toward new diff image\n",
        "              old_frame = do_3d_step('oldFrameScaled.png', frame_num, midas_model, midas_transform)\n",
        "              old_frame.save('oldFrameScaled.png')\n",
        "              if frame_num % int(turbo_steps) != 0: \n",
        "                print('turbo skip this frame: skipping clip diffusion steps')\n",
        "                filename = f'{args.batch_name}({args.batchNum})_{frame_num:04}.png'\n",
        "                blend_factor = ((frame_num % int(turbo_steps))+1)/int(turbo_steps)\n",
        "                print('turbo skip this frame: skipping clip diffusion steps and saving blended frame')\n",
        "                newWarpedImg = cv2.imread('prevFrameScaled.png')#this is already updated..\n",
        "                oldWarpedImg = cv2.imread('oldFrameScaled.png')\n",
        "                blendedImage = cv2.addWeighted(newWarpedImg, blend_factor, oldWarpedImg,1-blend_factor, 0.0)\n",
        "                cv2.imwrite(f'{batchFolder}/{filename}',blendedImage)\n",
        "                next_step_pil.save(f'{img_filepath}') # save it also as prev_frame to feed next iteration\n",
        "                if vr_mode:\n",
        "                  generate_eye_views(TRANSLATION_SCALE,batchFolder,filename,frame_num,midas_model, midas_transform)\n",
        "                continue\n",
        "              else:\n",
        "                #if not a skip frame, will run diffusion and need to blend.\n",
        "                oldWarpedImg = cv2.imread('prevFrameScaled.png')\n",
        "                cv2.imwrite(f'oldFrameScaled.png',oldWarpedImg)#swap in for blending later \n",
        "                print('clip/diff this frame - generate clip diff image')\n",
        "\n",
        "          init_image = 'prevFrameScaled.png'\n",
        "          init_scale = args.frames_scale\n",
        "          skip_steps = args.calc_frames_skip_steps\n",
        "\n",
        "      if  args.animation_mode == \"Video Input\":\n",
        "        init_scale = args.video_init_frames_scale\n",
        "        skip_steps = args.calc_frames_skip_steps\n",
        "        if not video_init_seed_continuity:\n",
        "          seed += 1\n",
        "        if video_init_flow_warp:\n",
        "          if frame_num == 0: \n",
        "            skip_steps = args.video_init_skip_steps\n",
        "            init_image = f'{videoFramesFolder}/{frame_num+1:04}.jpg'\n",
        "          if frame_num > 0: \n",
        "            prev = PIL.Image.open(batchFolder+f\"/{batch_name}({batchNum})_{frame_num-1:04}.png\")\n",
        "            \n",
        "            frame1_path = f'{videoFramesFolder}/{frame_num:04}.jpg'\n",
        "            frame2 = PIL.Image.open(f'{videoFramesFolder}/{frame_num+1:04}.jpg')\n",
        "            flo_path = f\"/{flo_folder}/{frame1_path.split('/')[-1]}.npy\"\n",
        "            \n",
        "            init_image = 'warped.png'\n",
        "            print(video_init_flow_blend)\n",
        "            weights_path = None\n",
        "            if video_init_check_consistency:\n",
        "                # TBD\n",
        "                pass\n",
        " \n",
        "            warp(prev, frame2, flo_path, blend=video_init_flow_blend, weights_path=weights_path).save(init_image)\n",
        "            \n",
        "        else:\n",
        "          init_image = f'{videoFramesFolder}/{frame_num+1:04}.jpg'\n",
        "\n",
        "\n",
        "      loss_values = []\n",
        "  \n",
        "      if seed is not None:\n",
        "          np.random.seed(seed)\n",
        "          random.seed(seed)\n",
        "          torch.manual_seed(seed)\n",
        "          torch.cuda.manual_seed_all(seed)\n",
        "          torch.backends.cudnn.deterministic = True\n",
        "  \n",
        "      target_embeds, weights = [], []\n",
        "      \n",
        "      if args.prompts_series is not None and frame_num >= len(args.prompts_series):\n",
        "        frame_prompt = args.prompts_series[-1]\n",
        "      elif args.prompts_series is not None:\n",
        "        frame_prompt = args.prompts_series[frame_num]\n",
        "      else:\n",
        "        frame_prompt = []\n",
        "      \n",
        "      print(args.image_prompts_series)\n",
        "      if args.image_prompts_series is not None and frame_num >= len(args.image_prompts_series):\n",
        "        image_prompt = args.image_prompts_series[-1]\n",
        "      elif args.image_prompts_series is not None:\n",
        "        image_prompt = args.image_prompts_series[frame_num]\n",
        "      else:\n",
        "        image_prompt = []\n",
        "\n",
        "      print(f'Frame {frame_num} Prompt: {frame_prompt}')\n",
        "\n",
        "      model_stats = []\n",
        "      for clip_model in clip_models:\n",
        "            cutn = 16\n",
        "            model_stat = {\"clip_model\":None,\"target_embeds\":[],\"make_cutouts\":None,\"weights\":[]}\n",
        "            model_stat[\"clip_model\"] = clip_model\n",
        "            \n",
        "            for prompt in frame_prompt:\n",
        "                txt, weight = parse_prompt(prompt)\n",
        "                txt = clip_model.encode_text(clip.tokenize(prompt).to(device)).float()\n",
        "                \n",
        "                if args.fuzzy_prompt:\n",
        "                    for i in range(25):\n",
        "                        model_stat[\"target_embeds\"].append((txt + torch.randn(txt.shape).cuda() * args.rand_mag).clamp(0,1))\n",
        "                        model_stat[\"weights\"].append(weight)\n",
        "                else:\n",
        "                    model_stat[\"target_embeds\"].append(txt)\n",
        "                    model_stat[\"weights\"].append(weight)\n",
        "        \n",
        "            if image_prompt:\n",
        "              model_stat[\"make_cutouts\"] = MakeCutouts(clip_model.visual.input_resolution, cutn, skip_augs=skip_augs) \n",
        "              for prompt in image_prompt:\n",
        "                  path, weight = parse_prompt(prompt)\n",
        "                  img = Image.open(fetch(path)).convert('RGB')\n",
        "                  img = TF.resize(img, min(side_x, side_y, *img.size), T.InterpolationMode.LANCZOS)\n",
        "                  batch = model_stat[\"make_cutouts\"](TF.to_tensor(img).to(device).unsqueeze(0).mul(2).sub(1))\n",
        "                  embed = clip_model.encode_image(normalize(batch)).float()\n",
        "                  if fuzzy_prompt:\n",
        "                      for i in range(25):\n",
        "                          model_stat[\"target_embeds\"].append((embed + torch.randn(embed.shape).cuda() * rand_mag).clamp(0,1))\n",
        "                          weights.extend([weight / cutn] * cutn)\n",
        "                  else:\n",
        "                      model_stat[\"target_embeds\"].append(embed)\n",
        "                      model_stat[\"weights\"].extend([weight / cutn] * cutn)\n",
        "        \n",
        "            model_stat[\"target_embeds\"] = torch.cat(model_stat[\"target_embeds\"])\n",
        "            model_stats.append(model_stat)\n",
        "  \n",
        "      init = None\n",
        "      if init_image is not None:\n",
        "          init = Image.open(fetch(init_image)).convert('RGB')\n",
        "          init = init.resize((args.side_x, args.side_y), Image.LANCZOS)\n",
        "          init = TF.to_tensor(init).to(device).unsqueeze(0).mul(2).sub(1)\n",
        "      \n",
        "      if args.perlin_init:\n",
        "          if args.perlin_mode == 'color':\n",
        "              init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, False)\n",
        "              init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, False)\n",
        "          elif args.perlin_mode == 'gray':\n",
        "            init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, True)\n",
        "            init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, True)\n",
        "          else:\n",
        "            init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, False)\n",
        "            init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, True)\n",
        "          # init = TF.to_tensor(init).add(TF.to_tensor(init2)).div(2).to(device)\n",
        "          init = TF.to_tensor(init).add(TF.to_tensor(init2)).div(2).to(device).unsqueeze(0).mul(2).sub(1)\n",
        "          del init2\n",
        "  \n",
        "      cur_t = None\n",
        "  \n",
        "      def cond_fn(x, t, y=None):\n",
        "          with torch.enable_grad():\n",
        "              x_is_NaN = False\n",
        "              x = x.detach().requires_grad_()\n",
        "              n = x.shape[0]\n",
        "              if use_secondary_model is True:\n",
        "                alpha = torch.tensor(diffusion.sqrt_alphas_cumprod[cur_t], device=device, dtype=torch.float32)\n",
        "                sigma = torch.tensor(diffusion.sqrt_one_minus_alphas_cumprod[cur_t], device=device, dtype=torch.float32)\n",
        "                cosine_t = alpha_sigma_to_t(alpha, sigma)\n",
        "                out = secondary_model(x, cosine_t[None].repeat([n])).pred\n",
        "                fac = diffusion.sqrt_one_minus_alphas_cumprod[cur_t]\n",
        "                x_in = out * fac + x * (1 - fac)\n",
        "                x_in_grad = torch.zeros_like(x_in)\n",
        "              else:\n",
        "                my_t = torch.ones([n], device=device, dtype=torch.long) * cur_t\n",
        "                out = diffusion.p_mean_variance(model, x, my_t, clip_denoised=False, model_kwargs={'y': y})\n",
        "                fac = diffusion.sqrt_one_minus_alphas_cumprod[cur_t]\n",
        "                x_in = out['pred_xstart'] * fac + x * (1 - fac)\n",
        "                x_in_grad = torch.zeros_like(x_in)\n",
        "              for model_stat in model_stats:\n",
        "                for i in range(args.cutn_batches):\n",
        "                    t_int = int(t.item())+1 #errors on last step without +1, need to find source\n",
        "                    #when using SLIP Base model the dimensions need to be hard coded to avoid AttributeError: 'VisionTransformer' object has no attribute 'input_resolution'\n",
        "                    try:\n",
        "                        input_resolution=model_stat[\"clip_model\"].visual.input_resolution\n",
        "                    except:\n",
        "                        input_resolution=224\n",
        "\n",
        "                    cuts = MakeCutoutsDango(input_resolution,\n",
        "                            Overview= args.cut_overview[1000-t_int], \n",
        "                            InnerCrop = args.cut_innercut[1000-t_int],\n",
        "                            IC_Size_Pow=args.cut_ic_pow[1000-t_int],\n",
        "                            IC_Grey_P = args.cut_icgray_p[1000-t_int]\n",
        "                            )\n",
        "                    clip_in = normalize(cuts(x_in.add(1).div(2)))\n",
        "                    image_embeds = model_stat[\"clip_model\"].encode_image(clip_in).float()\n",
        "                    dists = spherical_dist_loss(image_embeds.unsqueeze(1), model_stat[\"target_embeds\"].unsqueeze(0))\n",
        "                    dists = spherical_dist_loss(image_embeds.unsqueeze(1), model_stat[\"target_embeds\"].unsqueeze(0))\n",
        "                    dists = dists.view([args.cut_overview[1000-t_int]+args.cut_innercut[1000-t_int], n, -1])\n",
        "\n",
        "                    step_weights = []\n",
        "                    for i in range(0, len(model_stat[\"weights\"])):\n",
        "                      step_weights += [model_stat[\"weights\"][i][1000-t_int]]\n",
        "                      \n",
        "                    if np.absolute(np.sum(step_weights)) < 1e-3: raise RuntimeError('The weights must not sum to 0.')\n",
        "                    step_weights_divisor = np.absolute(np.sum(step_weights))\n",
        "                    step_weights = [item / step_weights_divisor for item in step_weights]\n",
        "                    step_weights = torch.tensor(step_weights, device=device)\n",
        "                    losses = dists.mul(step_weights).sum(2).mean(0)\n",
        "                    loss_values.append(losses.sum().item()) # log loss, probably shouldn't do per cutn_batch\n",
        "                    x_in_grad += torch.autograd.grad(losses.sum() * clip_guidance_scale, x_in)[0] / cutn_batches\n",
        "\n",
        "              #print(\"step_weights are \" + str(step_weights))\n",
        "              tv_losses = tv_loss(x_in)\n",
        "              if use_secondary_model is True:\n",
        "                range_losses = range_loss(out)\n",
        "              else:\n",
        "                range_losses = range_loss(out['pred_xstart'])\n",
        "              sat_losses = torch.abs(x_in - x_in.clamp(min=-1,max=1)).mean()\n",
        "              loss = tv_losses.sum() * tv_scale + range_losses.sum() * range_scale + sat_losses.sum() * sat_scale\n",
        "              if init is not None and init_scale:\n",
        "                  init_losses = lpips_model(x_in, init)\n",
        "                  loss = loss + init_losses.sum() * init_scale\n",
        "              x_in_grad += torch.autograd.grad(loss, x_in)[0]\n",
        "              if torch.isnan(x_in_grad).any()==False:\n",
        "                  grad = -torch.autograd.grad(x_in, x, x_in_grad)[0]\n",
        "              else:\n",
        "                # print(\"NaN'd\")\n",
        "                x_is_NaN = True\n",
        "                grad = torch.zeros_like(x)\n",
        "          if args.clamp_grad and x_is_NaN == False:\n",
        "              magnitude = grad.square().mean().sqrt()\n",
        "              return grad * magnitude.clamp(max=args.clamp_max) / magnitude  #min=-0.02, min=-clamp_max, \n",
        "          return grad\n",
        "  \n",
        "      if args.diffusion_sampling_mode == 'ddim':\n",
        "          sample_fn = diffusion.ddim_sample_loop_progressive\n",
        "      else:\n",
        "          sample_fn = diffusion.plms_sample_loop_progressive\n",
        "\n",
        "\n",
        "      image_display = Output()\n",
        "      for i in range(args.n_batches):\n",
        "          if args.animation_mode == 'None':\n",
        "            display.clear_output(wait=True)\n",
        "            batchBar = tqdm(range(args.n_batches), desc =\"Batches\")\n",
        "            batchBar.n = i\n",
        "            batchBar.refresh()\n",
        "          print('')\n",
        "          display.display(image_display)\n",
        "          gc.collect()\n",
        "          torch.cuda.empty_cache()\n",
        "          cur_t = diffusion.num_timesteps - skip_steps - 1\n",
        "          total_steps = cur_t\n",
        "\n",
        "          if perlin_init:\n",
        "              init = regen_perlin()\n",
        "\n",
        "          if args.diffusion_sampling_mode == 'ddim':\n",
        "              samples = sample_fn(\n",
        "                  model,\n",
        "                  (batch_size, 3, args.side_y, args.side_x),\n",
        "                  clip_denoised=clip_denoised,\n",
        "                  model_kwargs={},\n",
        "                  cond_fn=cond_fn,\n",
        "                  progress=True,\n",
        "                  skip_timesteps=skip_steps,\n",
        "                  init_image=init,\n",
        "                  randomize_class=randomize_class,\n",
        "                  eta=eta,\n",
        "                  transformation_fn=symmetry_transformation_fn,\n",
        "                  transformation_percent=args.transformation_percent\n",
        "              )\n",
        "          else:\n",
        "              samples = sample_fn(\n",
        "                  model,\n",
        "                  (batch_size, 3, args.side_y, args.side_x),\n",
        "                  clip_denoised=clip_denoised,\n",
        "                  model_kwargs={},\n",
        "                  cond_fn=cond_fn,\n",
        "                  progress=True,\n",
        "                  skip_timesteps=skip_steps,\n",
        "                  init_image=init,\n",
        "                  randomize_class=randomize_class,\n",
        "                  order=2,\n",
        "              )\n",
        "          \n",
        "          \n",
        "          # with run_display:\n",
        "          # display.clear_output(wait=True)\n",
        "          for j, sample in enumerate(samples):    \n",
        "            cur_t -= 1\n",
        "            intermediateStep = False\n",
        "            if args.steps_per_checkpoint is not None:\n",
        "                if j % steps_per_checkpoint == 0 and j > 0:\n",
        "                  intermediateStep = True\n",
        "            elif j in args.intermediate_saves:\n",
        "              intermediateStep = True\n",
        "            with image_display:\n",
        "              if j % args.display_rate == 0 or cur_t == -1 or intermediateStep == True:\n",
        "                  for k, image in enumerate(sample['pred_xstart']):\n",
        "                      # tqdm.write(f'Batch {i}, step {j}, output {k}:')\n",
        "                      current_time = datetime.now().strftime('%y%m%d-%H%M%S_%f')\n",
        "                      percent = math.ceil(j/total_steps*100)\n",
        "                      if args.n_batches > 0:\n",
        "                        #if intermediates are saved to the subfolder, don't append a step or percentage to the name\n",
        "                        if cur_t == -1 and args.intermediates_in_subfolder is True:\n",
        "                          save_num = f'{frame_num:04}' if animation_mode != \"None\" else i\n",
        "                          filename = f'{args.batch_name}({args.batchNum})_{save_num}.png'\n",
        "                        else:\n",
        "                          #If we're working with percentages, append it\n",
        "                          if args.steps_per_checkpoint is not None:\n",
        "                            filename = f'{args.batch_name}({args.batchNum})_{i:04}-{percent:02}%.png'\n",
        "                          # Or else, iIf we're working with specific steps, append those\n",
        "                          else:\n",
        "                            filename = f'{args.batch_name}({args.batchNum})_{i:04}-{j:03}.png'\n",
        "                      image = TF.to_pil_image(image.add(1).div(2).clamp(0, 1))\n",
        "                      if j % args.display_rate == 0 or cur_t == -1:\n",
        "                        image.save('progress.png')\n",
        "                        display.clear_output(wait=True)\n",
        "                        display.display(display.Image('progress.png'))\n",
        "                      if args.steps_per_checkpoint is not None:\n",
        "                        if j % args.steps_per_checkpoint == 0 and j > 0:\n",
        "                          if args.intermediates_in_subfolder is True:\n",
        "                            image.save(f'{partialFolder}/{filename}')\n",
        "                          else:\n",
        "                            image.save(f'{batchFolder}/{filename}')\n",
        "                      else:\n",
        "                        if j in args.intermediate_saves:\n",
        "                          if args.intermediates_in_subfolder is True:\n",
        "                            image.save(f'{partialFolder}/{filename}')\n",
        "                          else:\n",
        "                            image.save(f'{batchFolder}/{filename}')\n",
        "                      if cur_t == -1:\n",
        "                        if frame_num == 0:\n",
        "                          save_settings()\n",
        "                        if args.animation_mode != \"None\":\n",
        "                          image.save('prevFrame.png')\n",
        "                        image.save(f'{batchFolder}/{filename}')\n",
        "                        if args.animation_mode == \"3D\":\n",
        "                          # If turbo, save a blended image\n",
        "                          if turbo_mode and frame_num > 0:\n",
        "                            # Mix new image with prevFrameScaled\n",
        "                            blend_factor = (1)/int(turbo_steps)\n",
        "                            newFrame = cv2.imread('prevFrame.png') # This is already updated..\n",
        "                            prev_frame_warped = cv2.imread('prevFrameScaled.png')\n",
        "                            blendedImage = cv2.addWeighted(newFrame, blend_factor, prev_frame_warped, (1-blend_factor), 0.0)\n",
        "                            cv2.imwrite(f'{batchFolder}/{filename}',blendedImage)\n",
        "                          else:\n",
        "                            image.save(f'{batchFolder}/{filename}')\n",
        "\n",
        "                          if vr_mode:\n",
        "                            generate_eye_views(TRANSLATION_SCALE, batchFolder, filename, frame_num, midas_model, midas_transform)\n",
        "\n",
        "                        # if frame_num != args.max_frames-1:\n",
        "                        #   display.clear_output()\n",
        "          \n",
        "          plt.plot(np.array(loss_values), 'r')\n",
        "\n",
        "def generate_eye_views(trans_scale,batchFolder,filename,frame_num,midas_model, midas_transform):\n",
        "   for i in range(2):\n",
        "      theta = vr_eye_angle * (math.pi/180)\n",
        "      ray_origin = math.cos(theta) * vr_ipd / 2 * (-1.0 if i==0 else 1.0)\n",
        "      ray_rotation = (theta if i==0 else -theta)\n",
        "      translate_xyz = [-(ray_origin)*trans_scale, 0,0]\n",
        "      rotate_xyz = [0, (ray_rotation), 0]\n",
        "      rot_mat = p3dT.euler_angles_to_matrix(torch.tensor(rotate_xyz, device=device), \"XYZ\").unsqueeze(0)\n",
        "      transformed_image = dxf.transform_image_3d(f'{batchFolder}/{filename}', midas_model, midas_transform, DEVICE,\n",
        "                                                      rot_mat, translate_xyz, args.near_plane, args.far_plane,\n",
        "                                                      args.fov, padding_mode=args.padding_mode,\n",
        "                                                      sampling_mode=args.sampling_mode, midas_weight=args.midas_weight,spherical=True)\n",
        "      eye_file_path = batchFolder+f\"/frame_{frame_num:04}\" + ('_l' if i==0 else '_r')+'.png'\n",
        "      transformed_image.save(eye_file_path)\n",
        "\n",
        "def save_settings():\n",
        "    setting_list = {\n",
        "      'text_prompts': text_prompts,\n",
        "      'image_prompts': image_prompts,\n",
        "      'clip_guidance_scale': clip_guidance_scale,\n",
        "      'tv_scale': tv_scale,\n",
        "      'range_scale': range_scale,\n",
        "      'sat_scale': sat_scale,\n",
        "      # 'cutn': cutn,\n",
        "      'cutn_batches': cutn_batches,\n",
        "      'max_frames': max_frames,\n",
        "      'interp_spline': interp_spline,\n",
        "      # 'rotation_per_frame': rotation_per_frame,\n",
        "      'init_image': init_image,\n",
        "      'init_scale': init_scale,\n",
        "      'skip_steps': skip_steps,\n",
        "      # 'zoom_per_frame': zoom_per_frame,\n",
        "      'frames_scale': frames_scale,\n",
        "      'frames_skip_steps': frames_skip_steps,\n",
        "      'perlin_init': perlin_init,\n",
        "      'perlin_mode': perlin_mode,\n",
        "      'skip_augs': skip_augs,\n",
        "      'randomize_class': randomize_class,\n",
        "      'clip_denoised': clip_denoised,\n",
        "      'clamp_grad': clamp_grad,\n",
        "      'clamp_max': clamp_max,\n",
        "      'seed': seed,\n",
        "      'fuzzy_prompt': fuzzy_prompt,\n",
        "      'rand_mag': rand_mag,\n",
        "      'eta': eta,\n",
        "      'width': width_height[0],\n",
        "      'height': width_height[1],\n",
        "      'diffusion_model': diffusion_model,\n",
        "      'use_secondary_model': use_secondary_model,\n",
        "      'steps': steps,\n",
        "      'diffusion_steps': diffusion_steps,\n",
        "      'diffusion_sampling_mode': diffusion_sampling_mode,\n",
        "      'ViTB32': ViTB32,\n",
        "      'ViTB16': ViTB16,\n",
        "      'ViTL14': ViTL14,\n",
        "      'ViTL14_336px': ViTL14_336px,\n",
        "      'RN101': RN101,\n",
        "      'RN50': RN50,\n",
        "      'RN50x4': RN50x4,\n",
        "      'RN50x16': RN50x16,\n",
        "      'RN50x64': RN50x64,\n",
        "      'ViTB32_laion2b_e16': ViTB32_laion2b_e16,\n",
        "      'ViTB32_laion400m_e31': ViTB32_laion400m_e31,\n",
        "      'ViTB32_laion400m_32': ViTB32_laion400m_32,\n",
        "      'ViTB32quickgelu_laion400m_e31': ViTB32quickgelu_laion400m_e31,\n",
        "      'ViTB32quickgelu_laion400m_e32': ViTB32quickgelu_laion400m_e32,\n",
        "      'ViTB16_laion400m_e31': ViTB16_laion400m_e31,\n",
        "      'ViTB16_laion400m_e32': ViTB16_laion400m_e32,\n",
        "      'RN50_yffcc15m': RN50_yffcc15m,\n",
        "      'RN50_cc12m': RN50_cc12m,\n",
        "      'RN50_quickgelu_yfcc15m': RN50_quickgelu_yfcc15m,\n",
        "      'RN50_quickgelu_cc12m': RN50_quickgelu_cc12m,\n",
        "      'RN101_yfcc15m': RN101_yfcc15m,\n",
        "      'RN101_quickgelu_yfcc15m': RN101_quickgelu_yfcc15m,\n",
        "      'cut_overview': str(cut_overview),\n",
        "      'cut_innercut': str(cut_innercut),\n",
        "      'cut_ic_pow': str(cut_ic_pow),\n",
        "      'cut_icgray_p': str(cut_icgray_p),\n",
        "      'key_frames': key_frames,\n",
        "      'max_frames': max_frames,\n",
        "      'angle': angle,\n",
        "      'zoom': zoom,\n",
        "      'translation_x': translation_x,\n",
        "      'translation_y': translation_y,\n",
        "      'translation_z': translation_z,\n",
        "      'rotation_3d_x': rotation_3d_x,\n",
        "      'rotation_3d_y': rotation_3d_y,\n",
        "      'rotation_3d_z': rotation_3d_z,\n",
        "      'midas_depth_model': midas_depth_model,\n",
        "      'midas_weight': midas_weight,\n",
        "      'near_plane': near_plane,\n",
        "      'far_plane': far_plane,\n",
        "      'fov': fov,\n",
        "      'padding_mode': padding_mode,\n",
        "      'sampling_mode': sampling_mode,\n",
        "      'video_init_path':video_init_path,\n",
        "      'extract_nth_frame':extract_nth_frame,\n",
        "      'video_init_seed_continuity': video_init_seed_continuity,\n",
        "      'turbo_mode':turbo_mode,\n",
        "      'turbo_steps':turbo_steps,\n",
        "      'turbo_preroll':turbo_preroll,\n",
        "      'use_horizontal_symmetry':use_horizontal_symmetry,\n",
        "      'use_vertical_symmetry':use_vertical_symmetry,\n",
        "      'transformation_percent':transformation_percent,\n",
        "      #video init settings\n",
        "      'video_init_steps': video_init_steps,\n",
        "      'video_init_clip_guidance_scale': video_init_clip_guidance_scale,\n",
        "      'video_init_tv_scale': video_init_tv_scale,\n",
        "      'video_init_range_scale': video_init_range_scale,\n",
        "      'video_init_sat_scale': video_init_sat_scale,\n",
        "      'video_init_cutn_batches': video_init_cutn_batches,\n",
        "      'video_init_skip_steps': video_init_skip_steps,\n",
        "      'video_init_frames_scale': video_init_frames_scale,\n",
        "      'video_init_frames_skip_steps': video_init_frames_skip_steps,\n",
        "      #warp settings\n",
        "      'video_init_flow_warp':video_init_flow_warp,\n",
        "      'video_init_flow_blend':video_init_flow_blend,\n",
        "      'video_init_check_consistency':video_init_check_consistency,\n",
        "      'video_init_blend_mode':video_init_blend_mode\n",
        "    }\n",
        "    # print('Settings:', setting_list)\n",
        "    with open(f\"{batchFolder}/{batch_name}({batchNum})_settings.txt\", \"w+\") as f:   #save settings\n",
        "        json.dump(setting_list, f, ensure_ascii=False, indent=4)\n",
        "\n",
        "# DEFINE SECONDARY MODEL\n",
        "\n",
        "#@title 1.6 Define the secondary diffusion model\n",
        "\n",
        "def append_dims(x, n):\n",
        "    return x[(Ellipsis, *(None,) * (n - x.ndim))]\n",
        "\n",
        "\n",
        "def expand_to_planes(x, shape):\n",
        "    return append_dims(x, len(shape)).repeat([1, 1, *shape[2:]])\n",
        "\n",
        "\n",
        "def alpha_sigma_to_t(alpha, sigma):\n",
        "    return torch.atan2(sigma, alpha) * 2 / math.pi\n",
        "\n",
        "\n",
        "def t_to_alpha_sigma(t):\n",
        "    return torch.cos(t * math.pi / 2), torch.sin(t * math.pi / 2)\n",
        "\n",
        "\n",
        "@dataclass\n",
        "class DiffusionOutput:\n",
        "    v: torch.Tensor\n",
        "    pred: torch.Tensor\n",
        "    eps: torch.Tensor\n",
        "\n",
        "\n",
        "class ConvBlock(nn.Sequential):\n",
        "    def __init__(self, c_in, c_out):\n",
        "        super().__init__(\n",
        "            nn.Conv2d(c_in, c_out, 3, padding=1),\n",
        "            nn.ReLU(inplace=True),\n",
        "        )\n",
        "\n",
        "\n",
        "class SkipBlock(nn.Module):\n",
        "    def __init__(self, main, skip=None):\n",
        "        super().__init__()\n",
        "        self.main = nn.Sequential(*main)\n",
        "        self.skip = skip if skip else nn.Identity()\n",
        "\n",
        "    def forward(self, input):\n",
        "        return torch.cat([self.main(input), self.skip(input)], dim=1)\n",
        "\n",
        "\n",
        "class FourierFeatures(nn.Module):\n",
        "    def __init__(self, in_features, out_features, std=1.):\n",
        "        super().__init__()\n",
        "        assert out_features % 2 == 0\n",
        "        self.weight = nn.Parameter(torch.randn([out_features // 2, in_features]) * std)\n",
        "\n",
        "    def forward(self, input):\n",
        "        f = 2 * math.pi * input @ self.weight.T\n",
        "        return torch.cat([f.cos(), f.sin()], dim=-1)\n",
        "\n",
        "\n",
        "class SecondaryDiffusionImageNet(nn.Module):\n",
        "    def __init__(self):\n",
        "        super().__init__()\n",
        "        c = 64  # The base channel count\n",
        "\n",
        "        self.timestep_embed = FourierFeatures(1, 16)\n",
        "\n",
        "        self.net = nn.Sequential(\n",
        "            ConvBlock(3 + 16, c),\n",
        "            ConvBlock(c, c),\n",
        "            SkipBlock([\n",
        "                nn.AvgPool2d(2),\n",
        "                ConvBlock(c, c * 2),\n",
        "                ConvBlock(c * 2, c * 2),\n",
        "                SkipBlock([\n",
        "                    nn.AvgPool2d(2),\n",
        "                    ConvBlock(c * 2, c * 4),\n",
        "                    ConvBlock(c * 4, c * 4),\n",
        "                    SkipBlock([\n",
        "                        nn.AvgPool2d(2),\n",
        "                        ConvBlock(c * 4, c * 8),\n",
        "                        ConvBlock(c * 8, c * 4),\n",
        "                        nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False),\n",
        "                    ]),\n",
        "                    ConvBlock(c * 8, c * 4),\n",
        "                    ConvBlock(c * 4, c * 2),\n",
        "                    nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False),\n",
        "                ]),\n",
        "                ConvBlock(c * 4, c * 2),\n",
        "                ConvBlock(c * 2, c),\n",
        "                nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False),\n",
        "            ]),\n",
        "            ConvBlock(c * 2, c),\n",
        "            nn.Conv2d(c, 3, 3, padding=1),\n",
        "        )\n",
        "\n",
        "    def forward(self, input, t):\n",
        "        timestep_embed = expand_to_planes(self.timestep_embed(t[:, None]), input.shape)\n",
        "        v = self.net(torch.cat([input, timestep_embed], dim=1))\n",
        "        alphas, sigmas = map(partial(append_dims, n=v.ndim), t_to_alpha_sigma(t))\n",
        "        pred = input * alphas - v * sigmas\n",
        "        eps = input * sigmas + v * alphas\n",
        "        return DiffusionOutput(v, pred, eps)\n",
        "\n",
        "\n",
        "class SecondaryDiffusionImageNet2(nn.Module):\n",
        "    def __init__(self):\n",
        "        super().__init__()\n",
        "        c = 64  # The base channel count\n",
        "        cs = [c, c * 2, c * 2, c * 4, c * 4, c * 8]\n",
        "\n",
        "        self.timestep_embed = FourierFeatures(1, 16)\n",
        "        self.down = nn.AvgPool2d(2)\n",
        "        self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False)\n",
        "\n",
        "        self.net = nn.Sequential(\n",
        "            ConvBlock(3 + 16, cs[0]),\n",
        "            ConvBlock(cs[0], cs[0]),\n",
        "            SkipBlock([\n",
        "                self.down,\n",
        "                ConvBlock(cs[0], cs[1]),\n",
        "                ConvBlock(cs[1], cs[1]),\n",
        "                SkipBlock([\n",
        "                    self.down,\n",
        "                    ConvBlock(cs[1], cs[2]),\n",
        "                    ConvBlock(cs[2], cs[2]),\n",
        "                    SkipBlock([\n",
        "                        self.down,\n",
        "                        ConvBlock(cs[2], cs[3]),\n",
        "                        ConvBlock(cs[3], cs[3]),\n",
        "                        SkipBlock([\n",
        "                            self.down,\n",
        "                            ConvBlock(cs[3], cs[4]),\n",
        "                            ConvBlock(cs[4], cs[4]),\n",
        "                            SkipBlock([\n",
        "                                self.down,\n",
        "                                ConvBlock(cs[4], cs[5]),\n",
        "                                ConvBlock(cs[5], cs[5]),\n",
        "                                ConvBlock(cs[5], cs[5]),\n",
        "                                ConvBlock(cs[5], cs[4]),\n",
        "                                self.up,\n",
        "                            ]),\n",
        "                            ConvBlock(cs[4] * 2, cs[4]),\n",
        "                            ConvBlock(cs[4], cs[3]),\n",
        "                            self.up,\n",
        "                        ]),\n",
        "                        ConvBlock(cs[3] * 2, cs[3]),\n",
        "                        ConvBlock(cs[3], cs[2]),\n",
        "                        self.up,\n",
        "                    ]),\n",
        "                    ConvBlock(cs[2] * 2, cs[2]),\n",
        "                    ConvBlock(cs[2], cs[1]),\n",
        "                    self.up,\n",
        "                ]),\n",
        "                ConvBlock(cs[1] * 2, cs[1]),\n",
        "                ConvBlock(cs[1], cs[0]),\n",
        "                self.up,\n",
        "            ]),\n",
        "            ConvBlock(cs[0] * 2, cs[0]),\n",
        "            nn.Conv2d(cs[0], 3, 3, padding=1),\n",
        "        )\n",
        "\n",
        "    def forward(self, input, t):\n",
        "        timestep_embed = expand_to_planes(self.timestep_embed(t[:, None]), input.shape)\n",
        "        v = self.net(torch.cat([input, timestep_embed], dim=1))\n",
        "        alphas, sigmas = map(partial(append_dims, n=v.ndim), t_to_alpha_sigma(t))\n",
        "        pred = input * alphas - v * sigmas\n",
        "        eps = input * sigmas + v * alphas\n",
        "        return DiffusionOutput(v, pred, eps)\n",
        "\n"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "source": [
        ""
      ],
      "metadata": {
        "id": "G30IZD4IBLDE"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@title 1.3 Load Disco Diffusion Settings File\n",
        "\n",
        "# LOAD SETTINGS - Implemented by WASasquatch\n",
        "#@markdown ##### **Load Settings from File**\n",
        "#@markdown By default a settings file overrides base diffusion settings, but if you wish, you can select from the following options to overwride more to form crucial parts of, or an entire settings file.\n",
        "settings_file_path = None #@param{type: 'string'}\n",
        "#@markdown Loading a settings file to bypass these settings\n",
        "load_prompts_and_inits = False #@param{type: 'boolean'}\n",
        "#@markdown Load text and image prompts, as well as their related settings*\n",
        "load_diffusion_models = True #@param{type: 'boolean'}\n",
        "#@markdown Load diffusion model settings from file\n",
        "load_CLIP_models = True #@param{type: 'boolean'}\n",
        "#@markdown Load CLIP model settings from file"
      ],
      "metadata": {
        "cellView": "form",
        "id": "TOPRdNXa7Dlu"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DiffClipSetTop"
      },
      "source": [
        "# 2. Diffusion and CLIP model settings"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ModelSettings",
        "cellView": "form"
      },
      "source": [
        "#@markdown ####**Models Settings (note: For pixel art, the best is pixelartdiffusion_expanded):**\n",
        "diffusion_model = \"concept_art_generator_v000-1_alpha\" #@param [\"256x256_diffusion_uncond\", \"512x512_diffusion_uncond_finetune_008100\", \"concept_art_generator_v000-1_alpha\", \"portrait_generator_v001\", \"portrait_generator_v002\", \"portrait_generator_v003\", \"portrait_generator_v004\", \"pixelartdiffusion_expanded\", \"pixel_art_diffusion_hard_256\", \"pixel_art_diffusion_soft_256\", \"pixelartdiffusion4k\", \"watercolordiffusion_2\", \"watercolordiffusion\", \"PulpSciFiDiffusion\", \"Lithography_Diffusion\", \"Medieval_Diffusion\", \"Liminal_Diffusion_v1\", \"Liminal_Diffusion_Source\", \"Floral_Diffusion\", \"FeiArt_Handpainted_CG_Diffusion\", \"Textile_Diffusion\", \"Isometric_Diffusion_Revrart512px\", \"Laproper_Diffusion_Deepspace_256\", \"Schnippi_Diffusion_512x512_V2\", \"custom\"]\n",
        "\n",
        "go_big_diffusion_model = \"Inherit\" #@param [\"Inherit\", \"256x256_diffusion_uncond\", \"512x512_diffusion_uncond_finetune_008100\", \"concept_art_generator_v000-1_alpha\", \"portrait_generator_v001\", \"portrait_generator_v002\", \"portrait_generator_v003\", \"portrait_generator_v004\", \"pixelartdiffusion_expanded\", \"pixel_art_diffusion_hard_256\", \"pixel_art_diffusion_soft_256\", \"pixelartdiffusion4k\", \"watercolordiffusion_2\", \"watercolordiffusion\", \"PulpSciFiDiffusion\", \"Lithography_Diffusion\", \"Medieval_Diffusion\", \"Liminal_Diffusion_v1\", \"Liminal_Diffusion_Source\", \"Floral_Diffusion\", \"FeiArt_Handpainted_CG_Diffusion\", \"Textile_Diffusion\", \"Isometric_Diffusion_Revrart512px\", \"Laproper_Diffusion_Deepspace_256\", \"Schnippi_Diffusion_512x512_V2\", \"custom\"]\n",
        "#@markdown `go_big_diffusion_model`: Leave on `Inherit` if you want Go BIG to use `diffusion_model` instead of a custom selection\n",
        "\n",
        "use_secondary_model = False #@param {type: 'boolean'}\n",
        "diffusion_sampling_mode = 'ddim' #@param ['plms','ddim']\n",
        "#@markdown #####**Custom model:**\n",
        "custom_path = '/content/drive/MyDrive/AI/Disco_Diffusion/models/ema_0.9999_070000.pt'#@param {type: 'string'}\n",
        "#@markdown <br>\n",
        "\n",
        "#@markdown ###**CLIP settings:**\n",
        "use_checkpoint = True #@param {type: 'boolean'}\n",
        "ViTB32 = True #@param{type:\"boolean\"}\n",
        "ViTB16 = False #@param{type:\"boolean\"}\n",
        "ViTL14 = True #@param{type:\"boolean\"}\n",
        "ViTL14_336px = False #@param{type:\"boolean\"}\n",
        "RN101 = False #@param{type:\"boolean\"}\n",
        "RN50 = False #@param{type:\"boolean\"}\n",
        "RN50x4 = True #@param{type:\"boolean\"}\n",
        "RN50x16 = False #@param{type:\"boolean\"}\n",
        "RN50x64 = False #@param{type:\"boolean\"}\n",
        "#@markdown <br>\n",
        "\n",
        "#@markdown ###**OpenCLIP settings:**\n",
        "ViTB32_laion2b_e16 = False #@param{type:\"boolean\"}\n",
        "ViTB32_laion400m_e31 = False #@param{type:\"boolean\"}\n",
        "ViTB32_laion400m_32 = False #@param{type:\"boolean\"}\n",
        "ViTB32quickgelu_laion400m_e31 = False #@param{type:\"boolean\"}\n",
        "ViTB32quickgelu_laion400m_e32 = False #@param{type:\"boolean\"}\n",
        "ViTB16_laion400m_e31 = False #@param{type:\"boolean\"}\n",
        "ViTB16_laion400m_e32 = False #@param{type:\"boolean\"}\n",
        "RN50_yffcc15m = False #@param{type:\"boolean\"}\n",
        "RN50_cc12m = False #@param{type:\"boolean\"}\n",
        "RN50_quickgelu_yfcc15m = False #@param{type:\"boolean\"}\n",
        "RN50_quickgelu_cc12m = False #@param{type:\"boolean\"}\n",
        "RN101_yfcc15m = False #@param{type:\"boolean\"}\n",
        "RN101_quickgelu_yfcc15m = False #@param{type:\"boolean\"}\n",
        "#@markdown <br>\n",
        "\n",
        "#@markdown **Note:** If you're having issues with model downloads, check this to compare SHA's:\n",
        "check_model_SHA = False #@param{type:\"boolean\"}\n",
        "\n",
        "# LOAD SETTINGS - Implemented by WASasquatch\n",
        "if settings_file_path != None and settings_file_path != '' and settings_file_path.lower() != 'none':\n",
        "  if exists(settings_file_path):\n",
        "    with open(settings_file_path) as json_file:\n",
        "      jsd = json.load(json_file)\n",
        "      if load_diffusion_models:\n",
        "        print('Loading diffusion model settings from file: '+settings_file_path)\n",
        "        if jsd.__contains__('diffusion_model'):\n",
        "          diffusion_model = jsd['diffusion_model']\n",
        "        if jsd.__contains__('use_secondary_model'):\n",
        "          use_secondary_model = jsd['use_secondary_model']\n",
        "      if load_CLIP_models:\n",
        "        if jsd.__contains__('ViTB32'):\n",
        "          ViTB32 = jsd['ViTB32']\n",
        "        if jsd.__contains__('ViTB16'):\n",
        "          ViTB16 = jsd['ViTB16']\n",
        "        if jsd.__contains__('ViTL14'):\n",
        "          ViTL14 = jsd['ViTL14']\n",
        "        if jsd.__contains__('ViTL14_336px'):\n",
        "          ViTL14_336px = jsd['ViTL14_336px']\n",
        "        if jsd.__contains__('RN101'):\n",
        "          RN101 = jsd['RN101']\n",
        "        if jsd.__contains__('RN50'):\n",
        "          RN50 = jsd['RN50']\n",
        "        if jsd.__contains__('RN50x4'):\n",
        "          RN50x4 = jsd['RN50x4']\n",
        "        if jsd.__contains__('RN50x16'):\n",
        "          RN50x16 = jsd['RN50x16']\n",
        "        if jsd.__contains__('RN50x64'):\n",
        "          RN50x64 = jsd['RN50x64']\n",
        "        if jsd.__contains__('ViTB32_laion2b_e16'):\n",
        "          ViTB32_laion2b_e16 = jsd['ViTB32_laion2b_e16']\n",
        "        if jsd.__contains__('ViTB32_laion400m_e31'):\n",
        "          ViTB32_laion400m_e31 = jsd['ViTB32_laion400m_e31']\n",
        "        if jsd.__contains__('ViTB32_laion400m_32'):\n",
        "          ViTB32_laion400m_32 = jsd['ViTB32_laion400m_32']\n",
        "        if jsd.__contains__('ViTB32quickgelu_laion400m_e31'):\n",
        "          ViTB32quickgelu_laion400m_e31 = jsd['ViTB32quickgelu_laion400m_e31']\n",
        "        if jsd.__contains__('ViTB32quickgelu_laion400m_e32'):\n",
        "          ViTB32quickgelu_laion400m_e32 = jsd['ViTB32quickgelu_laion400m_e32']\n",
        "        if jsd.__contains__('ViTB16_laion400m_e31'):\n",
        "          ViTB16_laion400m_e31 = jsd['ViTB16_laion400m_e31']\n",
        "        if jsd.__contains__('ViTB16_laion400m_e32'):\n",
        "          ViTB16_laion400m_e32 = jsd['ViTB16_laion400m_e32']\n",
        "        if jsd.__contains__('RN50_yffcc15m'):\n",
        "          RN50_yffcc15m = jsd['RN50_yffcc15m']\n",
        "        if jsd.__contains__('RN50_cc12m'):\n",
        "          RN50_cc12m = jsd['RN50_cc12m']\n",
        "        if jsd.__contains__('RN50_quickgelu_yfcc15m'):\n",
        "          RN50_quickgelu_yfcc15m = jsd['RN50_quickgelu_yfcc15m']\n",
        "        if jsd.__contains__('RN50_quickgelu_cc12m'):\n",
        "          RN50_quickgelu_cc12m = jsd['RN50_quickgelu_cc12m']\n",
        "        if jsd.__contains__('RN101_yfcc15m'):\n",
        "          RN101_yfcc15m = jsd['RN101_yfcc15m']\n",
        "        if jsd.__contains__('RN101_quickgelu_yfcc15m'):\n",
        "          RN101_quickgelu_yfcc15m = jsd['RN101_quickgelu_yfcc15m']\n",
        "\n",
        "diff_model_map = {\n",
        "    '256x256_diffusion_uncond': { 'downloaded': False, 'sha': 'a37c32fffd316cd494cf3f35b339936debdc1576dad13fe57c42399a5dbc78b1', 'uri_list': ['https://openaipublic.blob.core.windows.net/diffusion/jul-2021/256x256_diffusion_uncond.pt', 'https://www.dropbox.com/s/9tqnqo930mpnpcn/256x256_diffusion_uncond.pt'] },\n",
        "    '512x512_diffusion_uncond_finetune_008100': { 'downloaded': False, 'sha': '9c111ab89e214862b76e1fa6a1b3f1d329b1a88281885943d2cdbe357ad57648', 'uri_list': ['https://the-eye.eu/public/AI/models/512x512_diffusion_unconditional_ImageNet/512x512_diffusion_uncond_finetune_008100.pt', 'https://huggingface.co/lowlevelware/512x512_diffusion_unconditional_ImageNet/resolve/main/512x512_diffusion_uncond_finetune_008100.pt'] },\n",
        "\n",
        "    'concept_art_generator_v000-1_alpha': { 'downloaded': False, 'sha': '0c0394148b4fb56234baee533c36f17202b8f936f28427a957f342aa2f18d040', 'uri_list': ['https://huggingface.co/WAS/concept_art_generator/resolve/main/concept_art_generator_v000-1_alpha.pt'] },\n",
        "\n",
        "    'portrait_generator_v001': { 'downloaded': False, 'sha': 'b7e8c747af880d4480b6707006f1ace000b058dd0eac5bb13558ba3752d9b5b9', 'uri_list': ['https://huggingface.co/felipe3dartist/portrait_generator_v001/resolve/main/portrait_generator_v001_ema_0.9999_1MM.pt'] },\n",
        "    'portrait_generator_v002': { 'downloaded': False, 'sha': '3bc39e28fd9690dafbbee83cc08d089a3640eca8ed4d14280c4a9d342c56fd7f', 'uri_list': ['https://huggingface.co/WAS/portrait-diffusion/resolve/main/ema_0.9999_165000.pt'] },\n",
        "    'portrait_generator_v003': { 'downloaded': False, 'sha': 'c0e1739731efe682f6429f2f7f905104b602b415ef8bf0b507034cce6050e8e3', 'uri_list': ['https://huggingface.co/WAS/portrait-diffusion/resolve/main/ema_0.9999_430000.pt'] },\n",
        "    'portrait_generator_v004': { 'downloaded': False, 'sha': '09daed35b70670b8491f27ef172e2a62f08f3a78326aeb88d23cade1fbb878ab', 'uri_list': ['https://huggingface.co/WAS/portrait-diffusion/resolve/main/ema_0.9999_080000.pt'] },\n",
        "\n",
        "    'pixelartdiffusion_expanded': { 'downloaded': False, 'sha': 'a73b40556634034bf43b5a716b531b46fb1ab890634d854f5bcbbef56838739a', 'uri_list': ['https://huggingface.co/KaliYuga/PADexpanded/resolve/main/PADexpanded.pt'] },\n",
        "    'pixel_art_diffusion_hard_256': { 'downloaded': False, 'sha': 'be4a9de943ec06eef32c65a1008c60ad017723a4d35dc13169c66bb322234161', 'uri_list': ['https://huggingface.co/KaliYuga/pixel_art_diffusion_hard_256/resolve/main/pixel_art_diffusion_hard_256.pt'] },\n",
        "    'pixel_art_diffusion_soft_256': { 'downloaded': False, 'sha': 'd321590e46b679bf6def1f1914b47c89e762c76f19ab3e3392c8ca07c791039c', 'uri_list': ['https://huggingface.co/KaliYuga/pixel_art_diffusion_soft_256/resolve/main/pixel_art_diffusion_soft_256.pt'] },\n",
        "    'pixelartdiffusion4k': { 'downloaded': False, 'sha': 'a1ba4f13f6dabb72b1064f15d8ae504d98d6192ad343572cc416deda7cccac30', 'uri_list': ['https://huggingface.co/KaliYuga/pixelartdiffusion4k/resolve/main/pixelartdiffusion4k.pt'] },\n",
        "    'watercolordiffusion_2': { 'downloaded': False, 'sha': '49c281b6092c61c49b0f1f8da93af9b94be7e0c20c71e662e2aa26fee0e4b1a9', 'uri_list': ['https://huggingface.co/KaliYuga/watercolordiffusion_2/resolve/main/watercolordiffusion_2.pt'] },\n",
        "    'watercolordiffusion': { 'downloaded': False, 'sha': 'a3e6522f0c8f278f90788298d66383b11ac763dd5e0d62f8252c962c23950bd6', 'uri_list': ['https://huggingface.co/KaliYuga/watercolordiffusion/resolve/main/watercolordiffusion.pt'] },\n",
        "    'PulpSciFiDiffusion': { 'downloaded': False, 'sha': 'b79e62613b9f50b8a3173e5f61f0320c7dbb16efad42a92ec94d014f6e17337f', 'uri_list': ['https://huggingface.co/KaliYuga/PulpSciFiDiffusion/resolve/main/PulpSciFiDiffusion.pt'] },\n",
        " \n",
        "    'Liminal_Diffusion_v1': { 'downloaded': False, 'sha': '87c36b544a367fceb0ca127d0028cd8a6f6b6e069e529b22999259d69c14f042', 'uri_list': ['https://huggingface.co/BrainArtLabs/liminal_diffusion/resolve/main/liminal_diffusion_v1.pt'] },\n",
        "    'Liminal_Diffusion_Source': { 'downloaded': False, 'sha': 'ce0064b8cea56c8adb4e5aa0ee2d02f65cd8f1baa905cc6504f462f1aac6d6f4', 'uri_list': ['https://huggingface.co/BrainArtLabs/liminal_diffusion/resolve/main/liminal_diffusion_source.pt'] },\n",
        "    'Medieval_Diffusion': { 'downloaded': False, 'sha': '1b66a0c9749f88b2d9124af7a3ba5c12d2645ffcc33356326269ccda8643a01b', 'uri_list': ['https://huggingface.co/KaliYuga/medievaldiffusion/resolve/main/medievaldiffusion.pt'] },\n",
        "    'Lithography_Diffusion': { 'downloaded': False, 'sha': 'a3e6522f0c8f278f90788298d66383b11ac763dd5e0d62f8252c962c23950bd6', 'uri_list': ['https://huggingface.co/KaliYuga/lithographydiffusion/resolve/main/lithographydiffusion.pt'] },\n",
        "    'Floral_Diffusion': { 'downloaded': False, 'sha': '197e9068f1ca0248fd89f9d6ca1f6f851783e04fddd0cc5b0020bfa655ed3aeb', 'uri_list': ['https://huggingface.co/jags/floraldiffusion/resolve/main/floraldiffusion.pt'] },\n",
        "    'FeiArt_Handpainted_CG_Diffusion': { 'downloaded': False, 'sha': '85f95f0618f288476ffcec9f48160542ba626f655b3df963543388dcd059f86a', 'uri_list': ['https://huggingface.co/Feiart/FeiArt-Handpainted-CG-Diffusion/resolve/main/FeiArt-Handpainted-CG-Diffusion.pt'] },\n",
        "    'Textile_Diffusion': { 'downloaded': False, 'sha': '82aa9ac10c67a806929b5399f04b933ffaa98f1a2bb0c18103d6ccd8f5bd2dd4', 'uri_list': ['https://huggingface.co/KaliYuga/textilediffusion/resolve/main/textilediffusion.pt'] },\n",
        "    'Isometric_Diffusion_Revrart512px': { 'downloaded': False, 'sha': '649bb7d10ea5170b71bc24adfb17f0305955fde38d2bb1bc5428c6d5baf9811c', 'uri_list': ['https://huggingface.co/Revrart/IsometricDiffusionRevrart512px/resolve/main/IsometricDiffusionRevrart512px.pt'] },\n",
        "    'Laproper_Diffusion_Deepspace_256': { 'downloaded': False, 'sha': '7c0d742d714ee512edda3299b440f22c72e95fda3f32e24c4ba1f03e0c4b0524', 'uri_list': ['https://huggingface.co/laproper/diffusion-deepspace-256/resolve/main/ema_0.9999_102000.pt']},\n",
        "    'Schnippi_Diffusion_512x512_V2': { 'downloaded': False, 'sha': '9c111ab89e214862b76e1fa6a1b3f1d329b1a88281885943d2cdbe357ad57648', 'uri_list': ['https://huggingface.co/shnippi/shnippi_diffusion/resolve/main/512x512_shnippi_V2.pt']},\n",
        "\n",
        "    'secondary': { 'downloaded': False, 'sha': '983e3de6f95c88c81b2ca7ebb2c217933be1973b1ff058776b970f901584613a', 'uri_list': ['https://the-eye.eu/public/AI/models/v-diffusion/secondary_model_imagenet_2.pth', 'https://ipfs.pollinations.ai/ipfs/bafybeibaawhhk7fhyhvmm7x24zwwkeuocuizbqbcg5nqx64jq42j75rdiy/secondary_model_imagenet_2.pth'] },\n",
        "}\n",
        "\n",
        "# Why do we have these unused Kaliyuga lists?\n",
        "kaliyuga_pixel_art_model_names = ['pixelartdiffusion_expanded', 'pixel_art_diffusion_hard_256', 'pixel_art_diffusion_soft_256', 'pixelartdiffusion4k', 'PulpSciFiDiffusion']\n",
        "kaliyuga_watercolor_model_names = ['watercolordiffusion', 'watercolordiffusion_2']\n",
        "kaliyuga_pulpscifi_model_names = ['PulpSciFiDiffusion']\n",
        "diffusion_models_256x256_list = ['256x256_diffusion_uncond'] + kaliyuga_pixel_art_model_names + kaliyuga_watercolor_model_names + kaliyuga_pulpscifi_model_names\n",
        "\n",
        "# Model Functions\n",
        "\n",
        "def configurate_model(diffusion_model, use_checkpoint = False):\n",
        "    model_configurations = {\n",
        "      '512x512_diffusion_uncond_finetune_008100': {\n",
        "          'attention_resolutions': '32, 16, 8',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 250, \n",
        "          'image_size': 512,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 256,\n",
        "          'num_head_channels': 64,\n",
        "          'num_res_blocks': 2,\n",
        "          'resblock_updown': True,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': not useCPU,\n",
        "          'use_scale_shift_norm': True,\n",
        "      },\n",
        "      '256x256_diffusion_uncond': {\n",
        "          'attention_resolutions': '32, 16, 8',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000, \n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 250, \n",
        "          'image_size': 512,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 256,\n",
        "          'num_head_channels': 64,\n",
        "          'num_res_blocks': 2,\n",
        "          'resblock_updown': True,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': not useCPU,\n",
        "          'use_scale_shift_norm': True,\n",
        "      },\n",
        "      '512x512_OpenAI': {\n",
        "          'attention_resolutions': '32, 16, 8',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'image_size': 512,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 128,\n",
        "          'num_heads': 4,\n",
        "          'num_res_blocks': 2,\n",
        "          'resblock_updown': True,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': True,\n",
        "      },\n",
        "      'Liminal_Diffusion': {\n",
        "          'attention_resolutions': '16',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 256,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 128,\n",
        "          'num_heads': 1,\n",
        "          'num_res_blocks': 2,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': False,\n",
        "      },\n",
        "      'Lithography_Diffusion': {\n",
        "          'attention_resolutions': '16',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 256,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 128,\n",
        "          'num_heads': 1,\n",
        "          'num_res_blocks': 2,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': False,\n",
        "      },\n",
        "      'Medieval_Diffusion': {\n",
        "          'attention_resolutions': '16',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 256,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 128,\n",
        "          'num_heads': 1,\n",
        "          'num_res_blocks': 2,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': False,\n",
        "      },\n",
        "      'Floral_Diffusion': {\n",
        "          'attention_resolutions': '16',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 256,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 128,\n",
        "          'num_heads': 1,        \n",
        "          'num_res_blocks': 2,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': False,\n",
        "      },\n",
        "      'FeiArt_Handpainted_CG_Diffusion': {\n",
        "          'attention_resolutions': '32,16,8',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 512,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 256,\n",
        "          'num_head_channels': 64,\n",
        "          'num_res_blocks': 2,\n",
        "          'resblock_updown': True,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': True,\n",
        "      },\n",
        "      'Textile_Diffusion': {\n",
        "          'attention_resolutions': '16',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 256,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 128,\n",
        "          'num_heads': 1,\n",
        "          'num_res_blocks': 2,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': False,\n",
        "      },\n",
        "      'Isometric_Diffusion_Revrart512px': {\n",
        "          'attention_resolutions': '32, 16, 8',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 512,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 256,\n",
        "          'num_head_channels': 64,\n",
        "          'num_res_blocks': 2,\n",
        "          'resblock_updown': True,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': True,\n",
        "      }, \n",
        "      'Laproper_Diffusion_Deepspace_256': {\n",
        "          'attention_resolutions': '32, 16, 8',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 256,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 128,\n",
        "          'num_heads': 4,\n",
        "          'num_res_blocks': 2,\n",
        "          'resblock_updown': True,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': True,\n",
        "      },\n",
        "      'Schnippi_Diffusion_512x512_V2': {\n",
        "          'attention_resolutions': '32, 16, 8',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000, \n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 250, \n",
        "          'image_size': 512,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 256,\n",
        "          'num_head_channels': 64,\n",
        "          'num_res_blocks': 2,\n",
        "          'resblock_updown': True,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': not useCPU,\n",
        "          'use_scale_shift_norm': True,\n",
        "      },\n",
        "      'Kaliyuga': {\n",
        "          'attention_resolutions': '16',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 256,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 128,\n",
        "          'num_heads': 1,\n",
        "          'num_res_blocks': 2,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': False,\n",
        "        },\n",
        "    }\n",
        "\n",
        "    tmp_model_config = model_and_diffusion_defaults()\n",
        "\n",
        "    if diffusion_model == '512x512_diffusion_uncond_finetune_008100' or diffusion_model == 'Schnippi_Diffusion_512x512_V2':\n",
        "        tmp_model_config.update(model_configurations['512x512_diffusion_uncond_finetune_008100'])\n",
        "    elif diffusion_model == '256x256_diffusion_uncond':\n",
        "        tmp_model_config.update(model_configurations['256x256_diffusion_uncond'])\n",
        "    elif diffusion_model == 'concept_art_generator_v000-1_alpha'  or diffusion_model == 'portrait_generator_v001' or diffusion_model == 'portrait_generator_v002' or diffusion_model == 'portrait_generator_v003' or diffusion_model == 'portrait_generator_v004':\n",
        "        tmp_model_config.update(model_configurations['512x512_OpenAI'])\n",
        "    elif diffusion_model == 'Liminal_Diffusion_v1' or diffusion_model == 'liminal_diffusion_source': #256x256 model\n",
        "        tmp_model_config.update(model_configurations['Liminal_Diffusion'])\n",
        "    elif diffusion_model == 'Lithography_Diffusion': \n",
        "        tmp_model_config.update(model_configurations['Lithography_Diffusion'])\n",
        "    elif diffusion_model == 'Medieval_Diffusion': \n",
        "        tmp_model_config.update(model_configurations['Medieval_Diffusion'])\n",
        "    elif diffusion_model == 'Floral_Diffusion': \n",
        "        tmp_model_config.update(model_configurations['Floral_Diffusion'])\n",
        "    elif diffusion_model == 'FeiArt_Handpainted_CG_Diffusion':\n",
        "        tmp_model_config.update(model_configurations['FeiArt_Handpainted_CG_Diffusion'])\n",
        "    elif diffusion_model == 'Textile_Diffusion':\n",
        "        tmp_model_config.update(model_configurations['Textile_Diffusion'])\n",
        "    elif diffusion_model == 'Isometric_Diffusion_Revrart512px':\n",
        "        tmp_model_config.update(model_configurations['Isometric_Diffusion_Revrart512px'])\n",
        "    elif diffusion_model == 'Laproper_Diffusion_Deepspace_256':\n",
        "        tmp_model_config.update(model_configurations['Laproper_Diffusion_Deepspace_256'])\n",
        "    else:\n",
        "        tmp_model_config.update(model_configurations['Kaliyuga'])\n",
        "\n",
        "    return tmp_model_config\n",
        "\n",
        "def get_model_filename(diffusion_model_name):\n",
        "  model_uri = diff_model_map[diffusion_model_name]['uri_list'][0]\n",
        "  model_filename = os.path.basename(urlparse(model_uri).path)\n",
        "  return model_filename\n",
        "\n",
        "def check_sha_hash(diffusion_model_name, model_local_path):\n",
        "  with open(model_local_path, \"rb\") as f: return True if hashlib.sha256(f.read()).hexdigest() == diff_model_map[diffusion_model_name]['sha'] else False\n",
        "\n",
        "def download_model(diffusion_model_name, uri_index=0):\n",
        "  if diffusion_model_name is 'custom':\n",
        "    return\n",
        "  model_filename = get_model_filename(diffusion_model_name)\n",
        "  model_local_path = os.path.join(model_path, model_filename)\n",
        "  print(\"Downloading Diffusion Model: \", diffusion_model_name)\n",
        "  if os.path.exists(model_local_path):\n",
        "    if check_model_SHA:\n",
        "      print(f\"Checking {model_filename} SHA256 Hash...\")\n",
        "      if check_sha_hash(diffusion_model_name, model_local_path):\n",
        "        print(f'{diffusion_model_name} SHA256 matches.')\n",
        "        diff_model_map[diffusion_model_name]['downloaded'] = True\n",
        "      else:\n",
        "        print(f\"{diffusion_model_name} SHA doesn't match. Will redownload it...\")\n",
        "        os.remove(model_local_path)\n",
        "        return download_model(diffusion_model_name)\n",
        "    print(f\"{model_filename} \\033[32malready downloaded.\\033[0;0m\")\n",
        "  else:\n",
        "    for model_uri in diff_model_map[diffusion_model_name]['uri_list']:\n",
        "      print(f\"Downloading {diffusion_model_name} from: {model_uri}\")\n",
        "      !wget -q --show-progress --no-cache $model_uri -P $model_path\n",
        "      if os.path.exists(model_local_path):\n",
        "        if check_model_SHA:\n",
        "          print(f\"Checking {model_filename} SHA256 Hash...\")\n",
        "          if check_sha_hash(diffusion_model_name, model_local_path):\n",
        "            print(f'{diffusion_model_name} SHA256 matches.')\n",
        "            diff_model_map[diffusion_model_name]['downloaded'] = True\n",
        "            print(f\"{diffusion_model_name} \\033[32mdownloaded successfully.\\033[0;0m\")\n",
        "            return\n",
        "          else:\n",
        "            print(f\"{diffusion_model_name} \\033[31mSHA doesn't match!\\033[0;0m Attempting to redownload\")\n",
        "            os.remove(model_local_path)\n",
        "            return download_model(diffusion_model_name)\n",
        "        else:\n",
        "          diff_model_map[diffusion_model_name]['downloaded'] = True\n",
        "          print(f\"{diffusion_model_name} \\033[32mdownloaded successfully.\\033[0;0m\")\n",
        "          return\n",
        "      else:\n",
        "        print(f\"\\033[31mFailed to download\\033[0;0m {diffusion_model_name} from: {model_uri}\")\n",
        "        if len(diff_model_map[diffusion_model_name]['uri_list']) > 1:\n",
        "          print(\"Attempting to download from another source...\")\n",
        "        else:\n",
        "          print(f\"No additional download resources available for {diffusion_model_name}\")\n",
        "\n",
        "\n",
        "# Download the diffusion model(s)\n",
        "download_model(diffusion_model)\n",
        "if use_secondary_model:\n",
        "    download_model('secondary')\n",
        "if go_big_diffusion_model != \"Inherit\":\n",
        "    download_model(go_big_diffusion_model)\n",
        "\n",
        "# Configurate diffusion model\n",
        "model_config = configurate_model(diffusion_model, use_checkpoint)\n",
        "model_default = model_config['image_size']\n",
        "if go_big_diffusion_model != \"Inherit\":\n",
        "    go_big_model_config = configurate_model(go_big_diffusion_model, use_checkpoint)\n",
        "    go_big_model_default = go_big_model_config['image_size']\n",
        "\n",
        "if use_secondary_model:\n",
        "    secondary_model = SecondaryDiffusionImageNet2()\n",
        "    secondary_model.load_state_dict(torch.load(f'{model_path}/secondary_model_imagenet_2.pth', map_location='cpu'))\n",
        "    secondary_model.eval().requires_grad_(False).to(device)\n",
        "\n",
        "clip_models = []\n",
        "if ViTB32: print(\"Downloading CLIP Model: ViT-B/32\"); clip_models.append(clip.load('ViT-B/32', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if ViTB16: print(\"Downloading CLIP Model: ViT-B/16\"); clip_models.append(clip.load('ViT-B/16', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if ViTL14: print(\"Downloading CLIP Model: ViT-L/14\"); clip_models.append(clip.load('ViT-L/14', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if ViTL14_336px: print(\"Downloading CLIP Model: ViT-L/14@336px\"); clip_models.append(clip.load('ViT-L/14@336px', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if RN50: print(\"Downloading CLIP Model: RN50\"); clip_models.append(clip.load('RN50', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if RN50x4: print(\"Downloading CLIP Model: RN50x4\"); clip_models.append(clip.load('RN50x4', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if RN50x16: print(\"Downloading CLIP Model: RN50x16\"); clip_models.append(clip.load('RN50x16', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if RN50x64: print(\"Downloading CLIP Model: RN50x64\"); clip_models.append(clip.load('RN50x64', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if RN101: print(\"Downloading CLIP Model: RN101\");clip_models.append(clip.load('RN101', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "\n",
        "if ViTB32_laion2b_e16: print(\"Downloading CLIP Model: ViT-B/32 laion2b_e16\"); clip_models.append(open_clip.create_model('ViT-B-32', pretrained='laion2b_e16').eval().requires_grad_(False).to(device))\n",
        "if ViTB32_laion400m_e31: print(\"Downloading CLIP Model: ViT-B-32 laion400m_e31\"); clip_models.append(open_clip.create_model('ViT-B-32', pretrained='laion400m_e31').eval().requires_grad_(False).to(device))\n",
        "if ViTB32_laion400m_32: print(\"Downloading CLIP Model: ViT-B/32 laion400m_e32\"); clip_models.append(open_clip.create_model('ViT-B-32', pretrained='laion400m_e32').eval().requires_grad_(False).to(device))\n",
        "if ViTB32quickgelu_laion400m_e31: print(\"Downloading CLIP Model: ViT-B-32-quickgelu laion400m_e31\"); clip_models.append(open_clip.create_model('ViT-B-32-quickgelu', pretrained='laion400m_e31').eval().requires_grad_(False).to(device))\n",
        "if ViTB32quickgelu_laion400m_e32: print(\"Downloading CLIP Model: ViT-B-32-quickgelu laion400m_e32\"); clip_models.append(open_clip.create_model('ViT-B-32-quickgelu', pretrained='laion400m_e32').eval().requires_grad_(False).to(device))\n",
        "if ViTB16_laion400m_e31: print(\"Downloading CLIP Model: ViT-B-16 laion400m_e31\"); clip_models.append(open_clip.create_model('ViT-B-16', pretrained='laion400m_e31').eval().requires_grad_(False).to(device))\n",
        "if ViTB16_laion400m_e32: print(\"Downloading CLIP Model: ViT-B-16 laion400m_e32\"); clip_models.append(open_clip.create_model('ViT-B-16', pretrained='laion400m_e32').eval().requires_grad_(False).to(device))\n",
        "if RN50_yffcc15m: print(\"Downloading CLIP Model: RN50 yfcc15m\"); clip_models.append(open_clip.create_model('RN50', pretrained='yfcc15m').eval().requires_grad_(False).to(device))\n",
        "if RN50_cc12m: print(\"Downloading CLIP Model: RN50 yfcc15m\"); clip_models.append(open_clip.create_model('RN50', pretrained='cc12m').eval().requires_grad_(False).to(device))\n",
        "if RN50_quickgelu_yfcc15m: print(\"Downloading CLIP Model: RN50-quickgelu yfcc15m\"); clip_models.append(open_clip.create_model('RN50-quickgelu', pretrained='yfcc15m').eval().requires_grad_(False).to(device))\n",
        "if RN50_quickgelu_cc12m: print(\"Downloading CLIP Model: RN50-quickgelu cc12m\"); clip_models.append(open_clip.create_model('RN50-quickgelu', pretrained='cc12m').eval().requires_grad_(False).to(device))\n",
        "if RN101_yfcc15m: print(\"Downloading CLIP Model: RN101 yfcc15m\"); clip_models.append(open_clip.create_model('RN101', pretrained='yfcc15m').eval().requires_grad_(False).to(device))\n",
        "if RN101_quickgelu_yfcc15m: print(\"Downloading CLIP Model: RN101=quickgelu yfcc15m\"); clip_models.append(open_clip.create_model('RN101-quickgelu', pretrained='yfcc15m').eval().requires_grad_(False).to(device))\n",
        "\n",
        "normalize = T.Normalize(mean=[0.48145466, 0.4578275, 0.40821073], std=[0.26862954, 0.26130258, 0.27577711])\n",
        "lpips_model = lpips.LPIPS(net='vgg').to(device)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CustModelTop"
      },
      "source": [
        "## Custom model settings \n",
        "Modify in accordance with your training settings and run the cell"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CustModel",
        "cellView": "form"
      },
      "source": [
        "#@markdown ####**Custom Model Settings:**\n",
        "if diffusion_model == 'custom':\n",
        "  model_config.update({\n",
        "        'attention_resolutions': '32, 16, 8',\n",
        "        'class_cond': False,\n",
        "        'diffusion_steps': 1000,\n",
        "        'rescale_timesteps': True,\n",
        "        'image_size': 512,\n",
        "        'learn_sigma': True,\n",
        "        'noise_schedule': 'linear',\n",
        "        'num_channels': 128,\n",
        "        'num_heads': 4,\n",
        "        'num_res_blocks': 2,\n",
        "        'resblock_updown': True,\n",
        "        'use_checkpoint': use_checkpoint,\n",
        "        'use_fp16': True,\n",
        "        'use_scale_shift_norm': True,\n",
        "    })"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SettingsTop"
      },
      "source": [
        "# 3. Settings"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BasicSettings",
        "cellView": "form"
      },
      "source": [
        "#@markdown ##**Basic Settings:** <br><br>\n",
        "\n",
        "#@markdown ### **Basic Diffusion Settings** \n",
        "batch_name = 'TimeToDisco' #@param{type: 'string'}\n",
        "steps =  100#@param [25,50,100,150,250,500,1000]{type: 'raw', allow-input: true}\n",
        "width_height_for_512x512_models = [512,768] #@param{type: 'raw'}\n",
        "clip_guidance_scale =  5000#@param{type: 'number'}\n",
        "tv_scale = 0#@param{type: 'number'}\n",
        "range_scale = 150#@param{type: 'number'}\n",
        "sat_scale = 0#@param{type: 'number'}\n",
        "cutn_batches = 1#@param{type: 'number'}\n",
        "skip_augs = False#@param{type: 'boolean'}\n",
        "\n",
        "#@markdown ####**Image dimensions to be used for 256x256 models (e.g. pixelart models):**\n",
        "width_height_for_256x256_models = [512, 512] #@param{type: 'raw'}\n",
        "#@markdown --- \n",
        "#@markdown <br><br>\n",
        "\n",
        "#@markdown ### **Go <font color=\"00a000\">BIG</font> Settings**\n",
        "#@markdown #### <font size=\"4\" color=\"CC0000\">*WARNING*</font>*: go_big_scale is a multipler* - if you go above **4** you might be waiting a long time\n",
        "#@markdown **NOTE**: If you have go_big_init set it will start with that image and skip the initial render\n",
        "\n",
        "go_big = False#@param{type: 'boolean'}\n",
        "go_big_scale = 2#@param{type: 'number'}\n",
        "gobig_skip_percent = 85#@param{type: 'number'}\n",
        "go_big_init = None #@param{type: 'string'}\n",
        "\n",
        "#@markdown Example: /content/drive/MyDrive/AI/Disco_Diffusion/init_images/image.png\n",
        "\n",
        "go_big_init_skipresize = False#@param{type: 'boolean'}\n",
        "#@markdown Check the above if you already resized the image. Please make sure go_big_scale is set to the multiplier you used.\n",
        "\n",
        "render_mask = None#@param{type: 'string'}\n",
        "#@markdown render_mask is a grayscale image that tells us where to draw (white) and not draw (black).\n",
        "#@markdown useful when using an init for go_big to avoid adding unwanted details in empty areas\n",
        "\n",
        "#@markdown --- \n",
        "#@markdown <br><br>\n",
        "\n",
        "#@markdown ###**Video Init Basic Settings:**\n",
        "video_init_steps = 100 #@param [25,50,100,150,250,500,1000]{type: 'raw', allow-input: true}\n",
        "video_init_clip_guidance_scale = 1000 #@param{type: 'number'}\n",
        "video_init_tv_scale = 0.1#@param{type: 'number'}\n",
        "video_init_range_scale = 150#@param{type: 'number'}\n",
        "video_init_sat_scale = 300#@param{type: 'number'}\n",
        "video_init_cutn_batches = 4#@param{type: 'number'}\n",
        "video_init_skip_steps = 50 #@param{type: 'integer'}\n",
        "\n",
        "#@markdown ---\n",
        "#@markdown <br><br>\n",
        "\n",
        "#@markdown ###**Init Image Settings:**\n",
        "#@markdown If your init image is too complex, and creating too much chaos, or you just want to use it as a base form consider using [Portrait Generator `init_image` Prep](https://rebrand.ly/dd-image-prep)\n",
        "init_image = None #@param{type: 'string'}\n",
        "init_scale = 0 #@param{type: 'integer'}\n",
        "skip_steps =  0 #@param{type: 'integer'}\n",
        "#@markdown *Make sure you set skip_steps to ~50% of your steps if you want to use an init image.* \n",
        "\n",
        "width_height = width_height_for_256x256_models if diffusion_model in diffusion_models_256x256_list else width_height_for_512x512_models\n",
        "\n",
        "#Get corrected sizes\n",
        "side_x = (width_height[0]//64)*64;\n",
        "side_y = (width_height[1]//64)*64;\n",
        "if side_x != width_height[0] or side_y != width_height[1]:\n",
        "    print(f'Changing output size to {side_x}x{side_y}. Dimensions must by multiples of 64.')\n",
        "\n",
        "#Make folder for batch\n",
        "batchFolder = f'{outDirPath}/{batch_name}'\n",
        "createPath(batchFolder)\n",
        "\n",
        "# Go BIG Calculations\n",
        "if go_big == True:\n",
        "  if go_big_init == \"\" or go_big_init == \"None\" or go_big_init == None:\n",
        "    go_big_init = None\n",
        "\n",
        "  if go_big_init_skipresize == True and go_big_init == None:\n",
        "    go_big_init_skipresize = False\n",
        "    raise Exception('You enabled skipresize but did not provide a go_big_init image')\n",
        "\n",
        "  if go_big_init != None:\n",
        "    print(f'side_x was {side_x}, side_y was {side_y}')\n",
        "    print(f'width_height was {width_height}')\n",
        "    go_big_temp_image = Image.open(fetch(go_big_init)).convert('RGB')\n",
        "    side_x, side_y = go_big_temp_image.size\n",
        "    print(f'side_x is now {side_x}, side_y is now {side_y}')\n",
        "    width_height[0] = side_x\n",
        "    width_height[1] = side_y\n",
        "    print(f'width_height is now {width_height}')\n",
        "    #go_big_temp_image.close\n",
        "    side_x = (width_height[0]//64)*64;\n",
        "    side_y = (width_height[1]//64)*64;\n",
        "    print(f'Slice side_x is {side_x}, side_y is {side_y}')\n",
        "    if side_x != width_height[0] or side_y != width_height[1]:\n",
        "      print('ERROR: Your go big init resolution was NOT a multiple of 64.')\n",
        "      raise Exception('Exiting due to improperly sized go big init.')\n"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AnimSetTop"
      },
      "source": [
        "### Animation Settings"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AnimSettings",
        "cellView": "form"
      },
      "source": [
        "#@markdown ####**Animation Mode:**\n",
        "animation_mode = 'None' #@param ['None', '2D', '3D', 'Video Input'] {type:'string'}\n",
        "#@markdown *For animation, you probably want to turn `cutn_batches` to 1 to make it quicker.*\n",
        "\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ####**Video Input Settings:**\n",
        "if is_colab:\n",
        "    video_init_path = \"/content/drive/MyDrive/init.mp4\" #@param {type: 'string'}\n",
        "else:\n",
        "    video_init_path = \"init.mp4\" #@param {type: 'string'}\n",
        "extract_nth_frame = 2 #@param {type: 'number'}\n",
        "persistent_frame_output_in_batch_folder = True #@param {type: 'boolean'}\n",
        "video_init_seed_continuity = False #@param {type: 'boolean'}\n",
        "#@markdown #####**Video Optical Flow Settings:**\n",
        "video_init_flow_warp = True #@param {type: 'boolean'}\n",
        "# Call optical flow from video frames and warp prev frame with flow\n",
        "video_init_flow_blend =  0.999#@param {type: 'number'} #0 - take next frame, 1 - take prev warped frame\n",
        "video_init_check_consistency = False #Insert param here when ready\n",
        "video_init_blend_mode = \"optical flow\" #@param ['None', 'linear', 'optical flow']\n",
        "# Call optical flow from video frames and warp prev frame with flow\n",
        "if animation_mode == \"Video Input\":\n",
        "    if persistent_frame_output_in_batch_folder or (not is_colab): #suggested by Chris the Wizard#8082 at discord\n",
        "        videoFramesFolder = f'{batchFolder}/videoFrames'\n",
        "    else:\n",
        "        videoFramesFolder = f'/content/videoFrames'\n",
        "    createPath(videoFramesFolder)\n",
        "    print(f\"Exporting Video Frames (1 every {extract_nth_frame})...\")\n",
        "    try:\n",
        "        for f in pathlib.Path(f'{videoFramesFolder}').glob('*.jpg'):\n",
        "            f.unlink()\n",
        "    except:\n",
        "        print('')\n",
        "    vf = f'select=not(mod(n\\,{extract_nth_frame}))'\n",
        "    if os.path.exists(video_init_path):\n",
        "        subprocess.run(['ffmpeg', '-i', f'{video_init_path}', '-vf', f'{vf}', '-vsync', 'vfr', '-q:v', '2', '-loglevel', 'error', '-stats', f'{videoFramesFolder}/%04d.jpg'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    else: \n",
        "        print(f'\\nWARNING!\\n\\nVideo not found: {video_init_path}.\\nPlease check your video path.\\n')\n",
        "    #!ffmpeg -i {video_init_path} -vf {vf} -vsync vfr -q:v 2 -loglevel error -stats {videoFramesFolder}/%04d.jpg\n",
        "\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ####**2D Animation Settings:**\n",
        "#@markdown `zoom` is a multiplier of dimensions, 1 is no zoom.\n",
        "#@markdown All rotations are provided in degrees.\n",
        "\n",
        "key_frames = True #@param {type:\"boolean\"}\n",
        "max_frames = 10000#@param {type:\"number\"}\n",
        "\n",
        "if animation_mode == \"Video Input\":\n",
        "    max_frames = len(glob(f'{videoFramesFolder}/*.jpg'))\n",
        "\n",
        "interp_spline = 'Linear' #Do not change, currently will not look good. param ['Linear','Quadratic','Cubic']{type:\"string\"}\n",
        "angle = \"0:(0)\"#@param {type:\"string\"}\n",
        "zoom = \"0: (1), 10: (1.05)\"#@param {type:\"string\"}\n",
        "translation_x = \"0: (0)\"#@param {type:\"string\"}\n",
        "translation_y = \"0: (0)\"#@param {type:\"string\"}\n",
        "translation_z = \"0: (10.0)\"#@param {type:\"string\"}\n",
        "rotation_3d_x = \"0: (0)\"#@param {type:\"string\"}\n",
        "rotation_3d_y = \"0: (0)\"#@param {type:\"string\"}\n",
        "rotation_3d_z = \"0: (0)\"#@param {type:\"string\"}\n",
        "midas_depth_model = \"dpt_large\"#@param {type:\"string\"}\n",
        "midas_weight = 0.3#@param {type:\"number\"}\n",
        "near_plane = 200#@param {type:\"number\"}\n",
        "far_plane = 10000#@param {type:\"number\"}\n",
        "fov = 40#@param {type:\"number\"}\n",
        "padding_mode = 'border'#@param {type:\"string\"}\n",
        "sampling_mode = 'bicubic'#@param {type:\"string\"}\n",
        "\n",
        "#======= TURBO MODE\n",
        "#@markdown ---\n",
        "#@markdown ####**Turbo Mode (3D anim only):**\n",
        "#@markdown (Starts after frame 10,) skips diffusion steps and just uses depth map to warp images for skipped frames.\n",
        "#@markdown Speeds up rendering by 2x-4x, and may improve image coherence between frames.\n",
        "#@markdown For different settings tuned for Turbo Mode, refer to the original Disco-Turbo Github: https://github.com/zippy731/disco-diffusion-turbo\n",
        "\n",
        "turbo_mode = False #@param {type:\"boolean\"}\n",
        "turbo_steps = \"3\" #@param [\"2\",\"3\",\"4\",\"5\",\"6\"] {type:\"string\"}\n",
        "turbo_preroll = 10 # frames\n",
        "\n",
        "#insist turbo be used only w 3d anim.\n",
        "if turbo_mode and animation_mode != '3D':\n",
        "    print('=====')\n",
        "    print('Turbo mode only available with 3D animations. Disabling Turbo.')\n",
        "    print('=====')\n",
        "    turbo_mode = False\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ####**Coherency Settings:**\n",
        "#@markdown `frame_scale` tries to guide the new frame to looking like the old one. A good default is 1500.\n",
        "frames_scale = 1500 #@param{type: 'integer'}\n",
        "#@markdown `frame_skip_steps` will blur the previous frame - higher values will flicker less but struggle to add enough new detail to zoom into.\n",
        "frames_skip_steps = '60%' #@param ['40%', '50%', '60%', '70%', '80%'] {type: 'string'}\n",
        "\n",
        "#@markdown ####**Video Init Coherency Settings:**\n",
        "#@markdown `frame_scale` tries to guide the new frame to looking like the old one. A good default is 1500.\n",
        "video_init_frames_scale = 15000 #@param{type: 'integer'}\n",
        "#@markdown `frame_skip_steps` will blur the previous frame - higher values will flicker less but struggle to add enough new detail to zoom into.\n",
        "video_init_frames_skip_steps = '70%' #@param ['40%', '50%', '60%', '70%', '80%'] {type: 'string'}\n",
        "\n",
        "#======= VR MODE\n",
        "#@markdown ---\n",
        "#@markdown ####**VR Mode (3D anim only):**\n",
        "#@markdown Enables stereo rendering of left/right eye views (supporting Turbo) which use a different (fish-eye) camera projection matrix.   \n",
        "#@markdown Note the images you're prompting will work better if they have some inherent wide-angle aspect\n",
        "#@markdown The generated images will need to be combined into left/right videos. These can then be stitched into the VR180 format.\n",
        "#@markdown Google made the VR180 Creator tool but subsequently stopped supporting it. It's available for download in a few places including https://www.patrickgrunwald.de/vr180-creator-download\n",
        "#@markdown The tool is not only good for stitching (videos and photos) but also for adding the correct metadata into existing videos, which is needed for services like YouTube to identify the format correctly.\n",
        "#@markdown Watching YouTube VR videos isn't necessarily the easiest depending on your headset. For instance Oculus have a dedicated media studio and store which makes the files easier to access on a Quest https://creator.oculus.com/manage/mediastudio/\n",
        "#@markdown \n",
        "#@markdown The command to get ffmpeg to concat your frames for each eye is in the form: `ffmpeg -framerate 15 -i frame_%4d_l.png l.mp4` (repeat for r)\n",
        "\n",
        "vr_mode = False #@param {type:\"boolean\"}\n",
        "#@markdown `vr_eye_angle` is the y-axis rotation of the eyes towards the center\n",
        "vr_eye_angle = 0.5 #@param{type:\"number\"}\n",
        "#@markdown interpupillary distance (between the eyes)\n",
        "vr_ipd = 5.0 #@param{type:\"number\"}\n",
        "\n",
        "#insist VR be used only w 3d anim.\n",
        "if vr_mode and animation_mode != '3D':\n",
        "    print('=====')\n",
        "    print('VR mode only available with 3D animations. Disabling VR.')\n",
        "    print('=====')\n",
        "    vr_mode = False\n",
        "\n",
        "\n",
        "def parse_key_frames(string, prompt_parser=None):\n",
        "    \"\"\"Given a string representing frame numbers paired with parameter values at that frame,\n",
        "    return a dictionary with the frame numbers as keys and the parameter values as the values.\n",
        "\n",
        "    Parameters\n",
        "    ----------\n",
        "    string: string\n",
        "        Frame numbers paired with parameter values at that frame number, in the format\n",
        "        'framenumber1: (parametervalues1), framenumber2: (parametervalues2), ...'\n",
        "    prompt_parser: function or None, optional\n",
        "        If provided, prompt_parser will be applied to each string of parameter values.\n",
        "    \n",
        "    Returns\n",
        "    -------\n",
        "    dict\n",
        "        Frame numbers as keys, parameter values at that frame number as values\n",
        "\n",
        "    Raises\n",
        "    ------\n",
        "    RuntimeError\n",
        "        If the input string does not match the expected format.\n",
        "    \n",
        "    Examples\n",
        "    --------\n",
        "    >>> parse_key_frames(\"10:(Apple: 1| Orange: 0), 20: (Apple: 0| Orange: 1| Peach: 1)\")\n",
        "    {10: 'Apple: 1| Orange: 0', 20: 'Apple: 0| Orange: 1| Peach: 1'}\n",
        "\n",
        "    >>> parse_key_frames(\"10:(Apple: 1| Orange: 0), 20: (Apple: 0| Orange: 1| Peach: 1)\", prompt_parser=lambda x: x.lower()))\n",
        "    {10: 'apple: 1| orange: 0', 20: 'apple: 0| orange: 1| peach: 1'}\n",
        "    \"\"\"\n",
        "    import re\n",
        "    pattern = r'((?P<frame>[0-9]+):[\\s]*[\\(](?P<param>[\\S\\s]*?)[\\)])'\n",
        "    frames = dict()\n",
        "    for match_object in re.finditer(pattern, string):\n",
        "        frame = int(match_object.groupdict()['frame'])\n",
        "        param = match_object.groupdict()['param']\n",
        "        if prompt_parser:\n",
        "            frames[frame] = prompt_parser(param)\n",
        "        else:\n",
        "            frames[frame] = param\n",
        "\n",
        "    if frames == {} and len(string) != 0:\n",
        "        raise RuntimeError('Key Frame string not correctly formatted')\n",
        "    return frames\n",
        "\n",
        "def get_inbetweens(key_frames, integer=False):\n",
        "    \"\"\"Given a dict with frame numbers as keys and a parameter value as values,\n",
        "    return a pandas Series containing the value of the parameter at every frame from 0 to max_frames.\n",
        "    Any values not provided in the input dict are calculated by linear interpolation between\n",
        "    the values of the previous and next provided frames. If there is no previous provided frame, then\n",
        "    the value is equal to the value of the next provided frame, or if there is no next provided frame,\n",
        "    then the value is equal to the value of the previous provided frame. If no frames are provided,\n",
        "    all frame values are NaN.\n",
        "\n",
        "    Parameters\n",
        "    ----------\n",
        "    key_frames: dict\n",
        "        A dict with integer frame numbers as keys and numerical values of a particular parameter as values.\n",
        "    integer: Bool, optional\n",
        "        If True, the values of the output series are converted to integers.\n",
        "        Otherwise, the values are floats.\n",
        "    \n",
        "    Returns\n",
        "    -------\n",
        "    pd.Series\n",
        "        A Series with length max_frames representing the parameter values for each frame.\n",
        "    \n",
        "    Examples\n",
        "    --------\n",
        "    >>> max_frames = 5\n",
        "    >>> get_inbetweens({1: 5, 3: 6})\n",
        "    0    5.0\n",
        "    1    5.0\n",
        "    2    5.5\n",
        "    3    6.0\n",
        "    4    6.0\n",
        "    dtype: float64\n",
        "\n",
        "    >>> get_inbetweens({1: 5, 3: 6}, integer=True)\n",
        "    0    5\n",
        "    1    5\n",
        "    2    5\n",
        "    3    6\n",
        "    4    6\n",
        "    dtype: int64\n",
        "    \"\"\"\n",
        "    key_frame_series = pd.Series([np.nan for a in range(max_frames)])\n",
        "\n",
        "    for i, value in key_frames.items():\n",
        "        key_frame_series[i] = value\n",
        "    key_frame_series = key_frame_series.astype(float)\n",
        "    \n",
        "    interp_method = interp_spline\n",
        "\n",
        "    if interp_method == 'Cubic' and len(key_frames.items()) <=3:\n",
        "      interp_method = 'Quadratic'\n",
        "    \n",
        "    if interp_method == 'Quadratic' and len(key_frames.items()) <= 2:\n",
        "      interp_method = 'Linear'\n",
        "      \n",
        "    \n",
        "    key_frame_series[0] = key_frame_series[key_frame_series.first_valid_index()]\n",
        "    key_frame_series[max_frames-1] = key_frame_series[key_frame_series.last_valid_index()]\n",
        "    # key_frame_series = key_frame_series.interpolate(method=intrp_method,order=1, limit_direction='both')\n",
        "    key_frame_series = key_frame_series.interpolate(method=interp_method.lower(),limit_direction='both')\n",
        "    if integer:\n",
        "        return key_frame_series.astype(int)\n",
        "    return key_frame_series\n",
        "\n",
        "def split_prompts(prompts):\n",
        "    prompt_series = pd.Series([np.nan for a in range(max_frames)])\n",
        "    for i, prompt in prompts.items():\n",
        "        prompt_series[i] = prompt\n",
        "    # prompt_series = prompt_series.astype(str)\n",
        "    prompt_series = prompt_series.ffill().bfill()\n",
        "    return prompt_series\n",
        "\n",
        "if key_frames:\n",
        "    try:\n",
        "        angle_series = get_inbetweens(parse_key_frames(angle))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `angle` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `angle` as \"\n",
        "            f'\"0: ({angle})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        angle = f\"0: ({angle})\"\n",
        "        angle_series = get_inbetweens(parse_key_frames(angle))\n",
        "\n",
        "    try:\n",
        "        zoom_series = get_inbetweens(parse_key_frames(zoom))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `zoom` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `zoom` as \"\n",
        "            f'\"0: ({zoom})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        zoom = f\"0: ({zoom})\"\n",
        "        zoom_series = get_inbetweens(parse_key_frames(zoom))\n",
        "\n",
        "    try:\n",
        "        translation_x_series = get_inbetweens(parse_key_frames(translation_x))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `translation_x` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `translation_x` as \"\n",
        "            f'\"0: ({translation_x})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        translation_x = f\"0: ({translation_x})\"\n",
        "        translation_x_series = get_inbetweens(parse_key_frames(translation_x))\n",
        "\n",
        "    try:\n",
        "        translation_y_series = get_inbetweens(parse_key_frames(translation_y))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `translation_y` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `translation_y` as \"\n",
        "            f'\"0: ({translation_y})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        translation_y = f\"0: ({translation_y})\"\n",
        "        translation_y_series = get_inbetweens(parse_key_frames(translation_y))\n",
        "\n",
        "    try:\n",
        "        translation_z_series = get_inbetweens(parse_key_frames(translation_z))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `translation_z` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `translation_z` as \"\n",
        "            f'\"0: ({translation_z})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        translation_z = f\"0: ({translation_z})\"\n",
        "        translation_z_series = get_inbetweens(parse_key_frames(translation_z))\n",
        "\n",
        "    try:\n",
        "        rotation_3d_x_series = get_inbetweens(parse_key_frames(rotation_3d_x))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `rotation_3d_x` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `rotation_3d_x` as \"\n",
        "            f'\"0: ({rotation_3d_x})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        rotation_3d_x = f\"0: ({rotation_3d_x})\"\n",
        "        rotation_3d_x_series = get_inbetweens(parse_key_frames(rotation_3d_x))\n",
        "\n",
        "    try:\n",
        "        rotation_3d_y_series = get_inbetweens(parse_key_frames(rotation_3d_y))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `rotation_3d_y` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `rotation_3d_y` as \"\n",
        "            f'\"0: ({rotation_3d_y})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        rotation_3d_y = f\"0: ({rotation_3d_y})\"\n",
        "        rotation_3d_y_series = get_inbetweens(parse_key_frames(rotation_3d_y))\n",
        "\n",
        "    try:\n",
        "        rotation_3d_z_series = get_inbetweens(parse_key_frames(rotation_3d_z))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `rotation_3d_z` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `rotation_3d_z` as \"\n",
        "            f'\"0: ({rotation_3d_z})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        rotation_3d_z = f\"0: ({rotation_3d_z})\"\n",
        "        rotation_3d_z_series = get_inbetweens(parse_key_frames(rotation_3d_z))\n",
        "\n",
        "else:\n",
        "    angle = float(angle)\n",
        "    zoom = float(zoom)\n",
        "    translation_x = float(translation_x)\n",
        "    translation_y = float(translation_y)\n",
        "    translation_z = float(translation_z)\n",
        "    rotation_3d_x = float(rotation_3d_x)\n",
        "    rotation_3d_y = float(rotation_3d_y)\n",
        "    rotation_3d_z = float(rotation_3d_z)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "InstallRAFT",
        "cellView": "form"
      },
      "source": [
        "#@title Install RAFT for Video input animation mode only\n",
        "#@markdown Run once per session. Doesn't download again if model path exists.\n",
        "#@markdown Use force download to reload raft models if needed\n",
        "force_download = False #@param {type:'boolean'}\n",
        "if animation_mode == 'Video Input':\n",
        "    try:\n",
        "        from raft import RAFT\n",
        "    except:\n",
        "        if not os.path.exists(os.path.join(PROJECT_DIR, 'RAFT')):\n",
        "            gitclone('https://github.com/princeton-vl/RAFT', os.path.join(PROJECT_DIR, 'RAFT'))\n",
        "        sys.path.append(f'{PROJECT_DIR}/RAFT')\n",
        "\n",
        "    if (not (os.path.exists(f'{root_path}/RAFT/models'))) or force_download:\n",
        "        createPath(f'{root_path}/RAFT')\n",
        "        os.chdir(f'{root_path}/RAFT')\n",
        "        sub_p_res = subprocess.run(['bash', f'{PROJECT_DIR}/RAFT/download_models.sh'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "        print(sub_p_res)\n",
        "        os.chdir(PROJECT_DIR)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FlowFns1",
        "cellView": "form"
      },
      "source": [
        "#@title Define optical flow functions for Video input animation mode only\n",
        "if animation_mode == 'Video Input':\n",
        "    in_path = videoFramesFolder\n",
        "    flo_folder = f'{in_path}/out_flo_fwd'\n",
        "    path = f'{PROJECT_DIR}/RAFT/core'\n",
        "    import sys\n",
        "    sys.path.append(f'{PROJECT_DIR}/RAFT/core')\n",
        "    os.chdir(f'{PROJECT_DIR}/RAFT/core')\n",
        "    print(os.getcwd())\n",
        "  \n",
        "    print(\"Renaming RAFT core's utils.utils to raftutils.utils (to avoid a naming conflict with AdaBins)\")\n",
        "    if not os.path.exists(f'{PROJECT_DIR}/RAFT/core/raftutils'):\n",
        "        os.rename(f'{PROJECT_DIR}/RAFT/core/utils', f'{PROJECT_DIR}/RAFT/core/raftutils')\n",
        "        sub_p_res = subprocess.run(['sed', '-i', 's/from utils.utils/from raftutils.utils/g', f'{PROJECT_DIR}/RAFT/core/corr.py'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "        sub_p_res = subprocess.run(['sed', '-i', 's/from utils.utils/from raftutils.utils/g', f'{PROJECT_DIR}/RAFT/core/raft.py'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "\n",
        "    from raftutils.utils import InputPadder\n",
        "    from raft import RAFT\n",
        "    from raftutils import flow_viz\n",
        "    import numpy as np\n",
        "    import argparse, PIL, cv2\n",
        "    from PIL import Image\n",
        "    from tqdm.notebook import tqdm\n",
        "    from glob import glob\n",
        "    import torch\n",
        "  \n",
        "    args2 = argparse.Namespace()\n",
        "    args2.small = False\n",
        "    args2.mixed_precision = True\n",
        "  \n",
        "  \n",
        "    TAG_CHAR = np.array([202021.25], np.float32)\n",
        "  \n",
        "    def writeFlow(filename,uv,v=None):\n",
        "        \"\"\" \n",
        "        https://github.com/NVIDIA/flownet2-pytorch/blob/master/utils/flow_utils.py\n",
        "        Copyright 2017 NVIDIA CORPORATION\n",
        "  \n",
        "        Licensed under the Apache License, Version 2.0 (the \"License\");\n",
        "        you may not use this file except in compliance with the License.\n",
        "        You may obtain a copy of the License at\n",
        "  \n",
        "            http://www.apache.org/licenses/LICENSE-2.0\n",
        "  \n",
        "        Unless required by applicable law or agreed to in writing, software\n",
        "        distributed under the License is distributed on an \"AS IS\" BASIS,\n",
        "        WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n",
        "        See the License for the specific language governing permissions and\n",
        "        limitations under the License.\n",
        "        \n",
        "        Write optical flow to file.\n",
        "        \n",
        "        If v is None, uv is assumed to contain both u and v channels,\n",
        "        stacked in depth.\n",
        "        Original code by Deqing Sun, adapted from Daniel Scharstein.\n",
        "        \"\"\"\n",
        "        nBands = 2\n",
        "  \n",
        "        if v is None:\n",
        "            assert(uv.ndim == 3)\n",
        "            assert(uv.shape[2] == 2)\n",
        "            u = uv[:,:,0]\n",
        "            v = uv[:,:,1]\n",
        "        else:\n",
        "            u = uv\n",
        "  \n",
        "        assert(u.shape == v.shape)\n",
        "        height,width = u.shape\n",
        "        f = open(filename,'wb')\n",
        "        # write the header\n",
        "        f.write(TAG_CHAR)\n",
        "        np.array(width).astype(np.int32).tofile(f)\n",
        "        np.array(height).astype(np.int32).tofile(f)\n",
        "        # arrange into matrix form\n",
        "        tmp = np.zeros((height, width*nBands))\n",
        "        tmp[:,np.arange(width)*2] = u\n",
        "        tmp[:,np.arange(width)*2 + 1] = v\n",
        "        tmp.astype(np.float32).tofile(f)\n",
        "        f.close()\n",
        "  \n",
        "    def load_img(img, size):\n",
        "        img = Image.open(img).convert('RGB').resize(size)\n",
        "        return torch.from_numpy(np.array(img)).permute(2,0,1).float()[None,...].cuda()\n",
        "  \n",
        "    def get_flow(frame1, frame2, model, iters=20):\n",
        "        padder = InputPadder(frame1.shape)\n",
        "        frame1, frame2 = padder.pad(frame1, frame2)\n",
        "        _, flow12 = model(frame1, frame2, iters=iters, test_mode=True)\n",
        "        flow12 = flow12[0].permute(1, 2, 0).detach().cpu().numpy()\n",
        "  \n",
        "        return flow12\n",
        "  \n",
        "    def warp_flow(img, flow):\n",
        "        h, w = flow.shape[:2]\n",
        "        flow = flow.copy()\n",
        "        flow[:, :, 0] += np.arange(w)\n",
        "        flow[:, :, 1] += np.arange(h)[:, np.newaxis]\n",
        "        res = cv2.remap(img, flow, None, cv2.INTER_LINEAR)\n",
        "        return res\n",
        "  \n",
        "    def makeEven(_x):\n",
        "        return _x if (_x % 2 == 0) else _x+1\n",
        "  \n",
        "    def fit(img,maxsize=512):\n",
        "        maxdim = max(*img.size)\n",
        "        if maxdim>maxsize:\n",
        "            # if True:\n",
        "            ratio = maxsize/maxdim\n",
        "            x,y = img.size\n",
        "            size = (makeEven(int(x*ratio)),makeEven(int(y*ratio))) \n",
        "            img = img.resize(size)\n",
        "        return img\n",
        "  \n",
        "    def warp(frame1, frame2, flo_path, blend=0.5, weights_path=None):\n",
        "        flow21 = np.load(flo_path)\n",
        "        frame1pil = np.array(frame1.convert('RGB').resize((flow21.shape[1],flow21.shape[0])))\n",
        "        frame1_warped21 = warp_flow(frame1pil, flow21)\n",
        "        # frame2pil = frame1pil\n",
        "        frame2pil = np.array(frame2.convert('RGB').resize((flow21.shape[1],flow21.shape[0])))\n",
        "    \n",
        "        if weights_path:\n",
        "            # TBD\n",
        "            pass\n",
        "        else:\n",
        "            blended_w = frame2pil*(1-blend) + frame1_warped21*(blend)\n",
        "  \n",
        "        return  PIL.Image.fromarray(blended_w.astype('uint8'))\n",
        "  \n",
        "    in_path = videoFramesFolder\n",
        "    flo_folder = f'{in_path}/out_flo_fwd'\n",
        "  \n",
        "    temp_flo = in_path+'/temp_flo'\n",
        "    flo_fwd_folder = in_path+'/out_flo_fwd'\n",
        "    # TBD flow backwards!\n",
        "  \n",
        "    os.chdir(PROJECT_DIR)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FlowFns2",
        "cellView": "form"
      },
      "source": [
        "#@title Generate optical flow and consistency maps\n",
        "#@markdown Run once per init video\n",
        "\n",
        "if animation_mode == \"Video Input\":\n",
        "    import gc\n",
        "\n",
        "    force_flow_generation = False #@param {type:'boolean'}\n",
        "    in_path = videoFramesFolder\n",
        "    flo_folder = f'{in_path}/out_flo_fwd'\n",
        "\n",
        "    if not video_init_flow_warp:\n",
        "        print('video_init_flow_warp not set, skipping')\n",
        "\n",
        "    if (animation_mode == 'Video Input') and (video_init_flow_warp):\n",
        "        flows = glob(flo_folder+'/*.*')\n",
        "        if (len(flows)>0) and not force_flow_generation:\n",
        "            print(f'Skipping flow generation:\\nFound {len(flows)} existing flow files in current working folder: {flo_folder}.\\nIf you wish to generate new flow files, check force_flow_generation and run this cell again.')\n",
        "    \n",
        "        if (len(flows)==0) or force_flow_generation:\n",
        "            frames = sorted(glob(in_path+'/*.*'));\n",
        "            if len(frames)<2: \n",
        "                print(f'WARNING!\\nCannot create flow maps: Found {len(frames)} frames extracted from your video input.\\nPlease check your video path.')\n",
        "            if len(frames)>=2:\n",
        "        \n",
        "                raft_model = torch.nn.DataParallel(RAFT(args2))\n",
        "                raft_model.load_state_dict(torch.load(f'{root_path}/RAFT/models/raft-things.pth'))\n",
        "                raft_model = raft_model.module.cuda().eval()\n",
        "        \n",
        "                for f in pathlib.Path(f'{flo_fwd_folder}').glob('*.*'):\n",
        "                    f.unlink()\n",
        "        \n",
        "                temp_flo = in_path+'/temp_flo'\n",
        "                flo_fwd_folder = in_path+'/out_flo_fwd'\n",
        "        \n",
        "                createPath(flo_fwd_folder)\n",
        "                createPath(temp_flo)\n",
        "        \n",
        "                # TBD Call out to a consistency checker?\n",
        "        \n",
        "                framecount = 0\n",
        "                for frame1, frame2 in tqdm(zip(frames[:-1], frames[1:]), total=len(frames)-1):\n",
        "        \n",
        "                    out_flow21_fn = f\"{flo_fwd_folder}/{frame1.split('/')[-1]}\"\n",
        "            \n",
        "                    frame1 = load_img(frame1, width_height)\n",
        "                    frame2 = load_img(frame2, width_height)\n",
        "            \n",
        "                    flow21 = get_flow(frame2, frame1, raft_model)\n",
        "                    np.save(out_flow21_fn, flow21)\n",
        "            \n",
        "                    if video_init_check_consistency:\n",
        "                        # TBD\n",
        "                        pass\n",
        "\n",
        "                del raft_model \n",
        "                gc.collect()"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ExtraSetTop"
      },
      "source": [
        "### Extra Settings\n",
        " Partial Saves, Advanced Settings, Cutn Scheduling"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ExtraSettings",
        "cellView": "form"
      },
      "source": [
        "#@markdown ###**Saving:**\n",
        "\n",
        "intermediate_saves = [80, 100, 135]#@param{type: 'raw'}\n",
        "intermediates_in_subfolder = False #@param{type: 'boolean'}\n",
        "#@markdown Intermediate steps will save a copy at your specified intervals. You can either format it as a single integer or a list of specific steps \n",
        "\n",
        "#@markdown A value of `2` will save a copy at 33% and 66%. 0 will save none.\n",
        "\n",
        "#@markdown A value of `[5, 9, 34, 45]` will save at steps 5, 9, 34, and 45. (Make sure to include the brackets)\n",
        "\n",
        "\n",
        "if type(intermediate_saves) is not list:\n",
        "    if intermediate_saves:\n",
        "        steps_per_checkpoint = math.floor((steps - skip_steps - 1) // (intermediate_saves+1))\n",
        "        steps_per_checkpoint = steps_per_checkpoint if steps_per_checkpoint > 0 else 1\n",
        "        print(f'Will save every {steps_per_checkpoint} steps')\n",
        "    else:\n",
        "        steps_per_checkpoint = steps+10\n",
        "else:\n",
        "    steps_per_checkpoint = None\n",
        "\n",
        "if intermediate_saves and intermediates_in_subfolder is True:\n",
        "    partialFolder = f'{batchFolder}/partials'\n",
        "    createPath(partialFolder)\n",
        "\n",
        "#@markdown <br>\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ###**Advanced Settings:**\n",
        "#@markdown *There are a few extra advanced settings available if you double click this cell.*\n",
        "\n",
        "#@markdown *Perlin init will replace your init, so uncheck if using one.*\n",
        "\n",
        "perlin_init = False  #@param{type: 'boolean'}\n",
        "perlin_mode = 'mixed' #@param ['mixed', 'color', 'gray']\n",
        "set_seed = 'random_seed' #@param{type: 'string'}\n",
        "eta = 0.8 #@param{type: 'number'}\n",
        "clamp_grad = True #@param{type: 'boolean'}\n",
        "clamp_max = 0.05 #@param{type: 'number'}\n",
        "\n",
        "#@markdown <br>\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "### EXTRA ADVANCED SETTINGS:\n",
        "#@markdown #### **Advanced Extra Settings:**\n",
        "randomize_class = True #@param{type: 'boolean'}\n",
        "clip_denoised = True #@param{type: 'boolean'}\n",
        "fuzzy_prompt = False #@param{type: 'boolean'}\n",
        "rand_mag = 0.05 #@param{type: 'number'}\n",
        "\n",
        "#@markdown <br>\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ### **Cutn Scheduling** (<font color=\"orange\">Easing</font>)\n",
        "#Add support for easing_functions\n",
        "try:\n",
        "  from easing_functions import *\n",
        "  import numpy as np\n",
        "except:\n",
        "  import subprocess\n",
        "  multipip_res = subprocess.run(['pip', 'install', 'easing_functions', 'matplotlib'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "  print(multipip_res)\n",
        "  from easing_functions import *\n",
        "  import numpy as np\n",
        "\n",
        "#@markdown To use easing, for cut_overview use the following format `[40]*100+dd_easing_fw(start=40, end=1, steps=800)+[1]*100`\n",
        "#@markdown and for cut_innercut use the following format `[1]*100+dd_easing_bw(start=1, end=40, steps=800)+[40]*100`\n",
        "#@markdown info on easing functions: https://github.com/semitable/easing-functions/blob/master/docs/cubic.png?raw=true\n",
        "#@markdown more info on easing functions: https://github.com/semitable/easing-functions\n",
        "#@markdown <br><br>\n",
        "#@markdown **Easing Function Aliases**\n",
        "#@markdown - Easing Integer: `easeint(st=40, en=1, s=800, fn=cut_easing_fn)` \n",
        "#@markdown - Easing Float: `easeflt(st=40, en=1, s=800, fn=cut_easing_fn)`\n",
        "#@markdown <br><br>\n",
        "\n",
        "#@markdown **cut_total_cuts**<br>\n",
        "#@markdown This can be used to specify the maximum number of cuts\n",
        "#@markdown which will then be distributed between overview and inner cuts. <br>\n",
        "#@markdown <br><br>\n",
        "\n",
        "def easing_normalize(arr, t_min, t_max):\n",
        "    norm_arr = []\n",
        "    for i in arr:\n",
        "        norm_arr.append((((i - min(arr))*(t_max - t_min))/(max(arr) - min(arr))) + t_min)\n",
        "    return norm_arr\n",
        "\n",
        "def dd_easing_int(start=1, end=9, steps=1000, easing_fn=\"ExponentialEaseInOut\"): return list(np.rint(easing_normalize(list(map(eval(easing_fn + \"(start=start, end=end, duration=steps)\"), np.arange(0, steps, 1))), min([start, end]), max([start, end]))).astype(int))\n",
        "def easeint(st=1, en=9, s=1000, fn=\"ExponentialEaseInOut\"): return dd_easing_int(start=en, end=en, steps=s, easing_fn=fn)\n",
        "def dd_easing_float(start=1, end=9, steps=1000, easing_fn=\"ExponentialEaseInOut\"): return list(easing_normalize(list(map(eval(easing_fn + \"(start=start, end=end, duration=steps)\"), np.arange(0, steps, 1))), min([start, end]), max([start, end])))\n",
        "def easeflt(st=1, en=9, s=1000, fn=\"ExponentialEaseInOut\"): return dd_easing_float(start=st, end=e, steps=s, easing_fn=fn)\n",
        "\n",
        "cut_easing_fn = \"ExponentialEaseInOut\" #@param [\"LinearInOut\", \"DEFAULT -> QuadEaseInOut\", \"CubicEaseInOut\", \"QuarticEaseInOut\", \"QuinticEaseInOut\", \"SineEaseInOut\", \"CircularEaseInOut\", \"ExponentialEaseInOut\", \"ElasticEaseInOut\", \"BackEaseInOut\", \"BounceEaseInOut\"]\n",
        "cut_total_cuts = 24 #@param{type: 'integer'}\n",
        "cut_overview = '[' + str(cut_total_cuts) + ']*200+dd_easing_int(start=' + str(cut_total_cuts) + ', end=' + str(cut_total_cuts/2) + ', steps=150, easing_fn=\"ExponentialEaseIn\")+dd_easing_int(start=' + str(cut_total_cuts/2) + ', end=0, steps=150, easing_fn=\"ExponentialEaseOut\")+[0]*500' #@param {type: 'string'}\n",
        "cut_overview_preset = \"None\" #@param [\"None\", \"DEFAULT -> [17]*400+dd_easing_int(start=1, end=17, steps=400, easing_fn=cut_easing_fn)+[0]*200\", \"[17]*200+dd_easing_int(start=1, end=19, steps=700, easing_fn=cut_easing_fn)+[3]*100\", \"DD DEFAULT -> [12]*400+[4]*600\"]\n",
        "cut_innercut = \"[\" + str(cut_total_cuts) + \"-x for x in eval(cut_overview)]\" #@param {type: 'string'}\n",
        "cut_innercut_preset = \"None\" #@param [\"None\", \"DEFAULT -> [1]*400+dd_easing_int(start=17, end=1, steps=400, easing_fn=cut_easing_fn)+[18]*200\", \"[3]*200+dd_easing_int(start=19, end=1, steps=700, easing_fn=cut_easing_fn)+[17]*100\", \"DD DEFAULT -> [4]*400+[12]*600\"]\n",
        "cut_ic_pow = '[0]*500+dd_easing_float(start=0, end=8, steps=300, easing_fn=cut_easing_fn)+[8]*200' #@param {type: 'string'}\n",
        "cut_ic_pow_preset = \"None\" #@param [\"None\", \"DEFAULT -> dd_easing_float(start=10, end=1, steps=1000, easing_fn=cut_easing_fn)\", \"[2]*500+dd_easing_float(start=10, end=1, steps=500, easing_fn=cut_easing_fn)\", \"[2.18]*500+dd_easing_float(start=10, end=1, steps=500, easing_fn=cut_easing_fn)\", \"dd_easing_float(start=1, end=8, steps=800, easing_fn=cut_easing_fn)+dd_easing_float(start=8, end=16, steps=200, easing_fn=cut_easing_fn)\", \"DD DEFAULT -> [1]*1000\", \"PORTRAIT DEFAULT -> [0.05]*1000\"]\n",
        "cut_icgray_p = \"[0.6]*100+[0.2]*100+[0.1]*100+[0.05]*100+[0]*600\" #@param {type: 'string'}\n",
        "cut_icgray_p_preset = \"None\" #@param [\"None\", \"DEFAULT -> [0.6]*100+[0.2]*100+[0.1]*100+[0.05]*100+[0]*600\", \"DD DEFAULT ->[0.2]*400+[0]*600\", \"[0.2]*400+[0.1]*600\"]\n",
        "\n",
        "#@markdown **Note:** Selecting a `_preset` will overwrite the customizable base value.\n",
        "\n",
        "skip_plot_view = False #@param{type: 'boolean'}\n",
        "#@markdown Skip using plot to display scheduling graph.\n",
        "\n",
        "# Overwrite settings with presets if selected\n",
        "if cut_overview_preset != \"None\": cut_overview = cut_overview_preset.split(' -> ')[1] if ' -> ' in cut_overview_preset else cut_overview_preset\n",
        "if cut_innercut_preset != \"None\": cut_innercut = cut_innercut_preset.split(' -> ')[1] if ' -> ' in cut_innercut_preset else cut_innercut_preset\n",
        "if cut_ic_pow_preset != \"None\": cut_ic_pow = cut_ic_pow_preset.split(' -> ')[1] if ' -> ' in cut_ic_pow_preset else cut_ic_pow_preset\n",
        "if cut_icgray_p_preset != \"None\": cut_icgray_p = cut_icgray_p.split(' -> ')[1] if ' -> ' in cut_icgray_p else cut_icgray_p\n",
        "\n",
        "# Scheduling Plotting\n",
        "\n",
        "if skip_plot_view == False:\n",
        "\n",
        "  import matplotlib.pyplot as plt\n",
        "\n",
        "  plt.plot(np.arange(0, 1000, 1), eval(cut_overview), label='overview cuts')\n",
        "  plt.plot(np.arange(0, 1000, 1), eval(cut_innercut), label='inner cuts')\n",
        "\n",
        "  total_cuts = []\n",
        "  cut_overview_evaled = eval(cut_overview)\n",
        "  cut_innercut_evaled = eval(cut_innercut)\n",
        "  for i in np.arange(0, 1000, 1):\n",
        "    total_cuts = total_cuts + [cut_overview_evaled[i] + cut_innercut_evaled[i]]\n",
        "\n",
        "  plt.plot(np.arange(0, 1000, 1), total_cuts, label='total cuts')\n",
        "  plt.plot(np.arange(0, 1000, 1), eval(cut_ic_pow), label='ic_pow')\n",
        "  plt.legend(loc='best')\n",
        "  plt.show()\n",
        "\n",
        "#@markdown <br>\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ### **KaliYuga Cutn Scheduling:**\n",
        "\n",
        "#@markdown KaliYuga model settings. Refer to [cut_ic_pow](https://ezcharts.miraheze.org/wiki/Category:Cut_ic_pow) as a guide. Values between 1 and 100 all work.\n",
        "pad_or_pulp_cut_overview = \"[15]*100+[15]*100+[12]*100+[12]*100+[6]*100+[4]*100+[2]*200+[0]*200\" #@param {type: 'string'}\n",
        "pad_or_pulp_cut_innercut = \"[1]*100+[1]*100+[4]*100+[4]*100+[8]*100+[8]*100+[10]*200+[10]*200\" #@param {type: 'string'}\n",
        "pad_or_pulp_cut_ic_pow = \"[12]*300+[12]*100+[12]*50+[12]*50+[10]*100+[10]*100+[10]*300\" #@param {type: 'string'}\n",
        "pad_or_pulp_cut_icgray_p = \"[0.87]*100+[0.78]*50+[0.73]*50+[0.64]*60+[0.56]*40+[0.50]*50+[0.33]*100+[0.19]*150+[0]*400\" #@param {type: 'string'}\n",
        "\n",
        "watercolor_cut_overview = \"[14]*200+[12]*200+[4]*400+[0]*200\" #@param {type: 'string'}\n",
        "watercolor_cut_innercut = \"[2]*200+[4]*200+[12]*400+[12]*200\" #@param {type: 'string'}\n",
        "watercolor_cut_ic_pow = \"[12]*300+[12]*100+[12]*50+[12]*50+[10]*100+[10]*100+[10]*300\" #@param {type: 'string'}\n",
        "watercolor_cut_icgray_p = \"[0.7]*100+[0.6]*100+[0.45]*100+[0.3]*100+[0]*600\" #@param {type: 'string'}\n",
        "\n",
        "if (diffusion_model in kaliyuga_pixel_art_model_names) or (diffusion_model in kaliyuga_pulpscifi_model_names):\n",
        "    cut_overview = pad_or_pulp_cut_overview\n",
        "    cut_innercut = pad_or_pulp_cut_innercut\n",
        "    cut_ic_pow = pad_or_pulp_cut_ic_pow\n",
        "    cut_icgray_p = pad_or_pulp_cut_icgray_p\n",
        "elif diffusion_model in kaliyuga_watercolor_model_names:\n",
        "    cut_overview = watercolor_cut_overview\n",
        "    cut_innercut = watercolor_cut_innercut\n",
        "    cut_ic_pow = watercolor_cut_ic_pow\n",
        "    cut_icgray_p = watercolor_cut_icgray_p\n",
        "\n",
        "#@markdown <br>\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ### **Symmetry Transformation Settings:**\n",
        "use_vertical_symmetry = False #@param {type:\"boolean\"}\n",
        "use_horizontal_symmetry = False #@param {type:\"boolean\"}\n",
        "transformation_percent = [0.09] #@param"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "source": [
        "# 4. Prompt Setup\n",
        "\n",
        "`animation_mode: None` will only use the first set. `animation_mode: 2D / Video` will run through them per the set frames and hold on to the last one.\n",
        "\n",
        "For help with Noodle Soup Prompts see the <a href=\"#reference_guide\">Reference section</a>"
      ],
      "metadata": {
        "id": "Ys8uUbuWc0bt"
      }
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Prompts"
      },
      "source": [
        "# Note: If using a pixelart diffusion model, try adding \"#pixelart\" to the end of the prompt for a stronger effect. It'll tend to work a lot better!\n",
        "# Note: Prompts now support scheduling. Ex: \"My awesome Disco Diffusion Prompt:dd_easing_float(start=15, end=0, steps=400, easing_fn=cut_easing_fn)\"\n",
        "\n",
        "text_prompts = {\n",
        "    0: [\n",
        "        \"Portrait of a Beautiful _adj-beauty_ _noun-emote_ Colonial Woman, trending on ArtStation, _site_.:5\",\n",
        "        \"Fine Art, Realistic, Detailed, Photorealistic, Realism Symmetrical Face:2\",\n",
        "        \"Photorealistic Color Scheme\",\n",
        "        \"Freckles, Blemishes, Moles:-1\",\n",
        "    ]\n",
        "}\n",
        "\n",
        "image_prompts = {\n",
        "    # 0:['ImagePromptsWorkButArentVeryGood.png:2',],\n",
        "}\n",
        "\n",
        "### PROMPT PREVIEW CODE - No need to edit\n",
        "\n",
        "#@title Preview Prompts\n",
        "use_this_prompt = True #@param{type: 'boolean'}\n",
        "#@markdown Use this random prompt instead of generating randomly on diffusion run. Note: The random prompt will be saved to the settings file.\n",
        "new_prompts = None\n",
        "if nsp: new_prompts = nsp_parse(text_prompts) \n",
        "if new_prompts: print(pprint.pformat(new_prompts))"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DiffuseTop"
      },
      "source": [
        "# 5. Diffuse!"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DoTheRun",
        "cellView": "form"
      },
      "source": [
        "#@title Do the Run!\n",
        "#@markdown `n_batches` ignored with animation modes.\n",
        "display_rate = 10 #@param{type: 'number'}\n",
        "n_batches = 5 #@param{type: 'number'}\n",
        "#@markdown **Note:** `n_batches` must be `1` for Go BIG. \n",
        "\n",
        "if gpu_warning:\n",
        "  print(gpu_warning+\"\\n\")\n",
        "\n",
        "if animation_mode == 'Video Input':\n",
        "    steps = video_init_steps\n",
        "\n",
        "#Update Model Settings\n",
        "timestep_respacing = f'ddim{steps}'\n",
        "diffusion_steps = (1000//steps)*steps if steps < 1000 else steps\n",
        "model_config.update({\n",
        "    'timestep_respacing': timestep_respacing,\n",
        "    'diffusion_steps': diffusion_steps,\n",
        "})\n",
        "\n",
        "batch_size = 1 \n",
        "\n",
        "def move_files(start_num, end_num, old_folder, new_folder):\n",
        "    for i in range(start_num, end_num):\n",
        "        old_file = old_folder + f'/{batch_name}({batchNum})_{i:04}.png'\n",
        "        new_file = new_folder + f'/{batch_name}({batchNum})_{i:04}.png'\n",
        "        os.rename(old_file, new_file)\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "\n",
        "resume_run = False #@param{type: 'boolean'}\n",
        "run_to_resume = 'latest' #@param{type: 'string'}\n",
        "resume_from_frame = 'latest' #@param{type: 'string'}\n",
        "retain_overwritten_frames = False #@param{type: 'boolean'}\n",
        "if retain_overwritten_frames:\n",
        "    retainFolder = f'{batchFolder}/retained'\n",
        "    createPath(retainFolder)\n",
        "\n",
        "\n",
        "# Parse Noodle Soup Prompts\n",
        "if nsp:\n",
        "  text_prompts = new_prompts if use_this_prompt else nsp_parse(text_prompts)\n",
        "\n",
        "\n",
        "skip_step_ratio = int(frames_skip_steps.rstrip(\"%\")) / 100\n",
        "calc_frames_skip_steps = math.floor(steps * skip_step_ratio)\n",
        "\n",
        "if animation_mode == 'Video Input':\n",
        "    frames = sorted(glob(in_path+'/*.*'));\n",
        "    if len(frames)==0: \n",
        "        sys.exit(\"ERROR: 0 frames found.\\nPlease check your video input path and rerun the video settings cell.\")\n",
        "    flows = glob(flo_folder+'/*.*')\n",
        "    if (len(flows)==0) and video_init_flow_warp:\n",
        "        sys.exit(\"ERROR: 0 flow files found.\\nPlease rerun the flow generation cell.\")\n",
        "\n",
        "if steps <= calc_frames_skip_steps:\n",
        "    sys.exit(\"ERROR: You can't skip more steps than your total steps\")\n",
        "\n",
        "if resume_run:\n",
        "    if run_to_resume == 'latest':\n",
        "        try:\n",
        "            batchNum\n",
        "        except:\n",
        "            batchNum = len(glob(f\"{batchFolder}/{batch_name}(*)_settings.txt\"))-1\n",
        "    else:\n",
        "        batchNum = int(run_to_resume)\n",
        "    if resume_from_frame == 'latest':\n",
        "        start_frame = len(glob(batchFolder+f\"/{batch_name}({batchNum})_*.png\"))\n",
        "        if animation_mode != '3D' and turbo_mode == True and start_frame > turbo_preroll and start_frame % int(turbo_steps) != 0:\n",
        "            start_frame = start_frame - (start_frame % int(turbo_steps))\n",
        "    else:\n",
        "        start_frame = int(resume_from_frame)+1\n",
        "        if animation_mode != '3D' and turbo_mode == True and start_frame > turbo_preroll and start_frame % int(turbo_steps) != 0:\n",
        "            start_frame = start_frame - (start_frame % int(turbo_steps))\n",
        "        if retain_overwritten_frames is True:\n",
        "            existing_frames = len(glob(batchFolder+f\"/{batch_name}({batchNum})_*.png\"))\n",
        "            frames_to_save = existing_frames - start_frame\n",
        "            print(f'Moving {frames_to_save} frames to the Retained folder')\n",
        "            move_files(start_frame, existing_frames, batchFolder, retainFolder)\n",
        "else:\n",
        "    start_frame = 0\n",
        "    batchNum = len(glob(batchFolder+\"/*.txt\"))\n",
        "    while os.path.isfile(f\"{batchFolder}/{batch_name}({batchNum})_settings.txt\") or os.path.isfile(f\"{batchFolder}/{batch_name}-{batchNum}_settings.txt\"):\n",
        "        batchNum += 1\n",
        "\n",
        "if go_big:\n",
        "  print(f'Starting Run: {batch_name}({batchNum}) at frame {start_frame} with Go BIG!')\n",
        "else:\n",
        "  print(f'Starting Run: {batch_name}({batchNum}) at frame {start_frame}')\n",
        "\n",
        "if set_seed == 'random_seed':\n",
        "    random.seed()\n",
        "    seed = random.randint(0, 2**32)\n",
        "    # print(f'Using seed: {seed}')\n",
        "else:\n",
        "    seed = int(set_seed)\n",
        "\n",
        "# LOAD SETTINGS - Implemented by WASasquatch\n",
        "if settings_file_path != None and settings_file_path != '' and settings_file_path.lower() != 'none':\n",
        "\n",
        "  import json\n",
        "  from os.path import exists\n",
        "\n",
        "  if exists(settings_file_path):\n",
        "\n",
        "    with open(settings_file_path) as json_file:\n",
        "\n",
        "      jsd = json.load(json_file)\n",
        "\n",
        "      if jsd:\n",
        "        print('Loading diffusion settings from file: '+settings_file_path)\n",
        "      if load_prompts_and_inits:\n",
        "        print('Loading text, image, and resolution prompt settings from file: '+settings_file_path)\n",
        "\n",
        "      if jsd.__contains__('clip_guidance_scale'):\n",
        "        clip_guidance_scale = jsd['clip_guidance_scale']\n",
        "      if jsd.__contains__('tv_scale'):\n",
        "        tv_scale = jsd['tv_scale']\n",
        "      if jsd.__contains__('range_scale'):\n",
        "        range_scale =  jsd['range_scale']\n",
        "      if jsd.__contains__('sat_scale'):\n",
        "        sat_scale = jsd['sat_scale']\n",
        "      if jsd.__contains__('cutn_batches'):\n",
        "        cutn_batches = jsd['cutn_batches']\n",
        "      if jsd.__contains__('interp_spline'):\n",
        "        interp_spline = jsd['interp_spline']\n",
        "      if jsd.__contains__('randomize_class'):\n",
        "        randomize_class = jsd['randomize_class']\n",
        "      if jsd.__contains__('clip_denoised'):\n",
        "        clip_denoised = jsd['clip_denoised']\n",
        "      if jsd.__contains__('clamp_gras'):\n",
        "        clamp_grad = jsd['clamp_gras']\n",
        "      if jsd.__contains__('clamp_max'):\n",
        "        clamp_max = jsd['clamp_max']\n",
        "      if jsd.__contains__('fuzzy_prompt'):\n",
        "        fuzzy_prompt = jsd['fuzzy_prompt']\n",
        "      if jsd.__contains__('max_frames'):\n",
        "        max_frames = jsd['max_frames']\n",
        "      if jsd.__contains__('rand_mag'):\n",
        "        rand_mag = jsd['rand_mag']\n",
        "      if jsd.__contains__('eta'):\n",
        "        eta = jsd['eta']\n",
        "      if jsd.__contains__('steps'):  \n",
        "        steps = jsd['steps']\n",
        "      if jsd.__contains__('diffusion_steps'):\n",
        "        diffusion_steps = jsd['diffusion_steps']\n",
        "      if jsd.__contains__('diffusion_sampling_mode'):\n",
        "        diffusion_sampling_mode = jsd['diffusion_sampling_mode']\n",
        "      if jsd.__contains__('cut_overview'):\n",
        "        cut_overview = jsd['cut_overview']\n",
        "      if jsd.__contains__('cut_innercut'):\n",
        "        cut_innercut = jsd['cut_innercut']\n",
        "      if jsd.__contains__('cut_ic_pow'):\n",
        "        cut_ic_pow = jsd['cut_ic_pow']\n",
        "      if jsd.__contains__('cut_icgray_p'):\n",
        "        cut_icgray_p = jsd['cut_icgray_p']\n",
        "      if jsd.__contains__('key_frames'):\n",
        "        key_frames = jsd['key_frames']\n",
        "      if jsd.__contains__('angle'):\n",
        "        angle = jsd['angle']\n",
        "      if jsd.__contains__('zoom'):\n",
        "        zoom = jsd['zoom']\n",
        "      if jsd.__contains__('translation_x'):\n",
        "        translation_x = jsd['translation_x']\n",
        "      if jsd.__contains__('translation_y'):\n",
        "        translation_y = jsd['translation_y']\n",
        "      if jsd.__contains__('translation_z'):\n",
        "        translation_z = jsd['translation_z']\n",
        "      if jsd.__contains__('translation_3d_x'):\n",
        "        translation_3d_x = jsd['translation_3d_x']\n",
        "      if jsd.__contains__('translation_3d_y'):\n",
        "        translation_3d_y = jsd['translation_3d_y']\n",
        "      if jsd.__contains__('translation_3d_z'):\n",
        "        translation_3d_z = jsd['translation_3d_z']\n",
        "      if jsd.__contains__('midas_depth_model'):\n",
        "        midas_depth_model = jsd['midas_depth_model']\n",
        "      if jsd.__contains__('midas_weight'):\n",
        "        midas_weight = jsd['midas_weight']\n",
        "      if jsd.__contains__('near_plane'):\n",
        "        near_plane = jsd['near_plane']\n",
        "      if jsd.__contains__('far_plane'):\n",
        "        far_plane = jsd['far_plane']\n",
        "      if jsd.__contains__('fov'):\n",
        "        fov = jsd['fov']\n",
        "      if jsd.__contains__('padding_mode'):\n",
        "        padding_mode = jsd['padding_mode']\n",
        "      if jsd.__contains__('sampling_mode'):\n",
        "        sampling_mode = jsd['sampling_mode']\n",
        "      if jsd.__contains__('extract_nth_frame'):\n",
        "        extract_nth_frame = jsd['extract_nth_frame']\n",
        "      if jsd.__contains__('turbo_mode'):\n",
        "        turbo_mode = jsd['turbo_mode']\n",
        "      if jsd.__contains__('turbo_steps'):\n",
        "        turbo_steps = jsd['turbo_steps']\n",
        "      if jsd.__contains__('turbo_preroll'):\n",
        "        turbo_preroll = jsd['turbo_preroll']\n",
        "      if jsd.__contains__('use_horizontal_symmetry'):\n",
        "        use_horizontal_symmetry = jsd['use_horizontal_symmetry']\n",
        "      if jsd.__contains__('use_vertical_symmetry'):\n",
        "        use_vertical_symmetry = jsd['use_vertical_symmetry']\n",
        "      if jsd.__contains__('transformation_percent'):\n",
        "        transformation_percent = jsd['transformation_percent']\n",
        "\n",
        "      if load_prompts_and_inits:\n",
        "\n",
        "        if jsd.__contains__('text_prompts'):\n",
        "          text_prompts = jsd['text_prompts']\n",
        "        if jsd.__contains__('image_prompts'):\n",
        "          image_prompts = jsd['image_prompts']\n",
        "        if jsd.__contains__('init_image'):\n",
        "          init_image = jsd['init_image']\n",
        "        if jsd.__contains__('init_scale'):\n",
        "          init_scale = jsd['init_scale']\n",
        "        if jsd.__contains__('skip_steps'):\n",
        "          skip_steps = jsd['skip_steps']\n",
        "        if jsd.__contains__('frames_scale'):\n",
        "          frames_scale = jsd['frames_scale']\n",
        "        if jsd.__contains__('frames_skip_steps'):\n",
        "          frames_skip_steps = jsd['frames_skip_steps']\n",
        "          skip_step_ratio = int(frames_skip_steps.rstrip(\"%\")) / 100\n",
        "          calc_frames_skip_steps = math.floor(steps * skip_step_ratio)\n",
        "        if jsd.__contains__('perlin_init'):\n",
        "          perlin_init = jsd['perlin_init']\n",
        "        if jsd.__contains__('perlin_mode'):\n",
        "          perlin_mode = jsd['perlin_mode']\n",
        "        if jsd.__contains__('width') and jsd.__contains__('height'):\n",
        "          width_height = [jsd['width'], jsd['height']]\n",
        "          side_x = jsd['width']\n",
        "          side_y = jsd['height']\n",
        "        if jsd.__contains__('video_init_seed_continuity'):\n",
        "          video_init_seed_continuity = jsd['video_init_seed_continuity']\n",
        "        if jsd.__contains__('video_init_steps'):\n",
        "          video_init_steps = jsd['video_init_steps']\n",
        "        if jsd.__contains__('video_init_clip_guidance_scale'):\n",
        "          video_init_clip_guidance_scale = jsd['video_init_clip_guidance_scale']\n",
        "        if jsd.__contains__('video_init_tv_scale'):\n",
        "          video_init_tv_scale = jsd['video_init_tv_scale']\n",
        "        if jsd.__contains__('video_init_tv_scale'):\n",
        "          video_init_range_scale = jsd['video_init_tv_scale']\n",
        "        if jsd.__contains__('video_init_sat_scale'):\n",
        "          video_init_sat_scale = jsd['video_init_sat_scale']\n",
        "        if jsd.__contains__('video_init_cutn_batches'):\n",
        "          video_init_cutn_batches = jsd['video_init_cutn_batches']\n",
        "        if jsd.__contains__('video_init_skip_steps'):\n",
        "          video_init_skip_steps = jsd['video_init_skip_steps']\n",
        "        if jsd.__contains__('video_init_frames_scale'):\n",
        "          video_init_frames_scale = jsd['video_init_frames_scale']\n",
        "        if jsd.__contains__('video_init_frames_skip_steps'):\n",
        "          video_init_frames_skip_steps = jsd['video_init_frames_skip_steps']\n",
        "        if jsd.__contains__('video_init_flow_warp'):\n",
        "          video_init_flow_warp = jsd['video_init_flow_warp']\n",
        "        if jsd.__contains__('video_init_flow_blend'):\n",
        "          video_init_flow_blend = jsd['video_init_flow_blend']\n",
        "        if jsd.__contains__('video_init_check_consistency'):\n",
        "          video_init_check_consistency = jsd['video_init_check_consistency']\n",
        "        if jsd.__contains__('video_init_blend_mode'):\n",
        "          video_init_blend_mode = jsd['video_init_blend_mode']\n",
        "\n",
        "# n_batches patch for Go BIG\n",
        "if go_big and n_batches > 1:\n",
        "  print(\"\\33[33m\\u001b[4mWARNING:\\33[0m n_batches is above 1, but Go BIG is active! Changing n_batches to 1...\")\n",
        "  n_batches = 1\n",
        "  import time\n",
        "  time.sleep(10)\n",
        "\n",
        "args = {\n",
        "    'batchNum': batchNum,\n",
        "    'prompts_series':split_prompts(text_prompts) if text_prompts else None,\n",
        "    'image_prompts_series':split_prompts(image_prompts) if image_prompts else None,\n",
        "    'seed': seed,\n",
        "    'display_rate':display_rate,\n",
        "    'n_batches':n_batches if animation_mode == 'None' else 1,\n",
        "    'batch_size':batch_size,\n",
        "    'batch_name': batch_name,\n",
        "    'steps': steps,\n",
        "    'diffusion_sampling_mode': diffusion_sampling_mode,\n",
        "    'width_height': width_height,\n",
        "    'clip_guidance_scale': clip_guidance_scale,\n",
        "    'tv_scale': tv_scale,\n",
        "    'range_scale': range_scale,\n",
        "    'sat_scale': sat_scale,\n",
        "    'cutn_batches': cutn_batches,\n",
        "    'init_image': init_image,\n",
        "    'init_scale': init_scale,\n",
        "    'skip_steps': skip_steps,\n",
        "    'side_x': side_x,\n",
        "    'side_y': side_y,\n",
        "    'timestep_respacing': timestep_respacing,\n",
        "    'diffusion_steps': diffusion_steps,\n",
        "    'animation_mode': animation_mode,\n",
        "    'video_init_path': video_init_path,\n",
        "    'extract_nth_frame': extract_nth_frame,\n",
        "    'video_init_seed_continuity': video_init_seed_continuity,\n",
        "    'key_frames': key_frames,\n",
        "    'max_frames': max_frames if animation_mode != \"None\" else 1,\n",
        "    'interp_spline': interp_spline,\n",
        "    'start_frame': start_frame,\n",
        "    'angle': angle,\n",
        "    'zoom': zoom,\n",
        "    'translation_x': translation_x,\n",
        "    'translation_y': translation_y,\n",
        "    'translation_z': translation_z,\n",
        "    'rotation_3d_x': rotation_3d_x,\n",
        "    'rotation_3d_y': rotation_3d_y,\n",
        "    'rotation_3d_z': rotation_3d_z,\n",
        "    'midas_depth_model': midas_depth_model,\n",
        "    'midas_weight': midas_weight,\n",
        "    'near_plane': near_plane,\n",
        "    'far_plane': far_plane,\n",
        "    'fov': fov,\n",
        "    'padding_mode': padding_mode,\n",
        "    'sampling_mode': sampling_mode,\n",
        "    'angle_series':angle_series,\n",
        "    'zoom_series':zoom_series,\n",
        "    'translation_x_series':translation_x_series,\n",
        "    'translation_y_series':translation_y_series,\n",
        "    'translation_z_series':translation_z_series,\n",
        "    'rotation_3d_x_series':rotation_3d_x_series,\n",
        "    'rotation_3d_y_series':rotation_3d_y_series,\n",
        "    'rotation_3d_z_series':rotation_3d_z_series,\n",
        "    'frames_scale': frames_scale,\n",
        "    'skip_step_ratio': skip_step_ratio,\n",
        "    'calc_frames_skip_steps': calc_frames_skip_steps,\n",
        "    'text_prompts': text_prompts,\n",
        "    'image_prompts': image_prompts,\n",
        "    'cut_overview': eval(cut_overview),\n",
        "    'cut_innercut': eval(cut_innercut),\n",
        "    'cut_ic_pow': eval(cut_ic_pow),\n",
        "    'cut_icgray_p': eval(cut_icgray_p),\n",
        "    'intermediate_saves': intermediate_saves,\n",
        "    'intermediates_in_subfolder': intermediates_in_subfolder,\n",
        "    'steps_per_checkpoint': steps_per_checkpoint,\n",
        "    'perlin_init': perlin_init,\n",
        "    'perlin_mode': perlin_mode,\n",
        "    'set_seed': set_seed,\n",
        "    'eta': eta,\n",
        "    'clamp_grad': clamp_grad,\n",
        "    'clamp_max': clamp_max,\n",
        "    'skip_augs': skip_augs,\n",
        "    'randomize_class': randomize_class,\n",
        "    'clip_denoised': clip_denoised,\n",
        "    'fuzzy_prompt': fuzzy_prompt,\n",
        "    'rand_mag': rand_mag,\n",
        "    'turbo_mode':turbo_mode,\n",
        "    'turbo_steps':turbo_steps,\n",
        "    'turbo_preroll':turbo_preroll,\n",
        "    'use_vertical_symmetry': use_vertical_symmetry,\n",
        "    'use_horizontal_symmetry': use_horizontal_symmetry,\n",
        "    'transformation_percent': transformation_percent,\n",
        "    #video init settings\n",
        "    'video_init_steps': video_init_steps,\n",
        "    'video_init_clip_guidance_scale': video_init_clip_guidance_scale,\n",
        "    'video_init_tv_scale': video_init_tv_scale,\n",
        "    'video_init_range_scale': video_init_range_scale,\n",
        "    'video_init_sat_scale': video_init_sat_scale,\n",
        "    'video_init_cutn_batches': video_init_cutn_batches,\n",
        "    'video_init_skip_steps': video_init_skip_steps,\n",
        "    'video_init_frames_scale': video_init_frames_scale,\n",
        "    'video_init_frames_skip_steps': video_init_frames_skip_steps,\n",
        "    #warp settings\n",
        "    'video_init_flow_warp':video_init_flow_warp,\n",
        "    'video_init_flow_blend':video_init_flow_blend,\n",
        "    'video_init_check_consistency':video_init_check_consistency,\n",
        "    'video_init_blend_mode':video_init_blend_mode\n",
        "}\n",
        "\n",
        "if animation_mode == 'Video Input':\n",
        "    # This isn't great in terms of what will get saved to the settings.. but it should work.\n",
        "    args['steps'] = args['video_init_steps']\n",
        "    args['clip_guidance_scale'] = args['video_init_clip_guidance_scale']\n",
        "    args['tv_scale'] = args['video_init_tv_scale']\n",
        "    args['range_scale'] = args['video_init_range_scale']\n",
        "    args['sat_scale'] = args['video_init_sat_scale']\n",
        "    args['cutn_batches'] = args['video_init_cutn_batches']\n",
        "    args['skip_steps'] = args['video_init_skip_steps']\n",
        "    args['frames_scale'] = args['video_init_frames_scale']\n",
        "    args['frames_skip_steps'] = args['video_init_frames_skip_steps']\n",
        "\n",
        "args = SimpleNamespace(**args)\n",
        "\n",
        "if go_big_init == None:\n",
        "\n",
        "  print('Preparing model...')\n",
        "  model, diffusion = create_model_and_diffusion(**model_config)\n",
        "  if diffusion_model == 'custom':\n",
        "      model.load_state_dict(torch.load(custom_path, map_location='cpu'))\n",
        "  else:\n",
        "      print(f'Using model: {get_model_filename(diffusion_model)}')\n",
        "      model.load_state_dict(torch.load(f'{model_path}/{get_model_filename(diffusion_model)}', map_location='cpu'))\n",
        "  model.requires_grad_(False).eval().to(device)\n",
        "  for name, param in model.named_parameters():\n",
        "      if 'qkv' in name or 'norm' in name or 'proj' in name:\n",
        "          param.requires_grad_()\n",
        "  if model_config['use_fp16']:\n",
        "      model.convert_to_fp16()\n",
        "\n",
        "\n",
        "# FUNCTIONS FOR GO BIG MODE\n",
        "slices_todo = (go_big_scale * go_big_scale) + 1\n",
        "if go_big_scale > 2:\n",
        "  slices_todo += 1 # add an extra slice to help ensure comically large images don't run out of memory\n",
        "\n",
        "# Input is an image, return image with mask added as an alpha channel\n",
        "def addalpha(im, mask):\n",
        "    imr, img, imb, ima = im.split()\n",
        "    mmr, mmg, mmb, mma = mask.split()\n",
        "    im = Image.merge('RGBA', [imr, img, imb, mma]) # we want the RGB from the original, but the transparency from the mask\n",
        "    return(im)\n",
        "\n",
        "# take a source image and layer in the slices on top\n",
        "def mergeimgs(source, slices):\n",
        "    source.convert(\"RGBA\")\n",
        "    width, height = source.size\n",
        "    slice_width, slice_height = slices[0].size\n",
        "    slice_width -= 64 # remove overlap\n",
        "    paste_x = 0\n",
        "    for slice in slices:\n",
        "        source.alpha_composite(slice, (paste_x,0))\n",
        "        paste_x += slice_width\n",
        "    return source\n",
        "\n",
        "# Slices an image into the configured number of chunks. Overlap is a quarter of the size of a chunk\n",
        "def slice(source, rmask):\n",
        "    global slices_todo\n",
        "    width, height = source.size\n",
        "    overlap = 64\n",
        "    slice_width = int(width / slices_todo)\n",
        "    slice_width = 64 * math.floor(slice_width / 64) #round slice width down to the nearest 64\n",
        "    if slice_width < 64:\n",
        "      slice_width = 64\n",
        "    remainder = width - (slice_width * slices_todo)\n",
        "    while remainder > 0:\n",
        "        slices_todo += 1\n",
        "        remainder = remainder - slice_width\n",
        "        print('Adding an extra slice to cover remainder of the image...')\n",
        "    slice_width += overlap\n",
        "    i = 0\n",
        "    slices = []\n",
        "    slice_rmasks = []\n",
        "    x = 0\n",
        "    y = 0\n",
        "    edgex = slice_width\n",
        "    while i < slices_todo:\n",
        "        slices.append(source.crop((x, y, edgex, height)))\n",
        "        slice_rmasks.append(rmask.crop((x, y, edgex, height)))\n",
        "        x += slice_width - overlap\n",
        "        edgex = x + slice_width\n",
        "        i += 1\n",
        "    slices_with_rmasks = zip(slices, slice_rmasks)\n",
        "    return (slices_with_rmasks)\n",
        "\n",
        "try:\n",
        "\n",
        "  if go_big_init is not None:\n",
        "\n",
        "    progress_image = 'progress.png'\n",
        "    go_big_temp_image.save(progress_image)\n",
        "    go_big_temp_image.close()\n",
        "    #shutil.copy(go_big_init, progress_image)\n",
        "\n",
        "  else:\n",
        "\n",
        "    do_run()\n",
        "\n",
        "  if go_big:\n",
        "\n",
        "    current_time = datetime.now().strftime('%y%m%d-%H%M%S_%f')\n",
        "    # Resize initial progress.png to new size\n",
        "    original_output_image = (f'{batchFolder}/{batch_name}_original_output_{current_time}.png')\n",
        "    final_output_image = (f'{batchFolder}/{batch_name}_final_output_{current_time}.png')\n",
        "    progress_image = 'progress.png'\n",
        "    input_image = Image.open(fetch(progress_image)).convert('RGBA')\n",
        "    input_image.save(original_output_image)\n",
        "    slice_image = (f'{batchFolder}/{batch_name}_slice.png')\n",
        "    slice_rmask = (f'{batchFolder}/{batch_name}_slice_rmask.png')\n",
        "\n",
        "    if render_mask is None:\n",
        "        source_render_mask = Image.new('RGBA', (args.side_x, args.side_y), color = (255,255,255))\n",
        "    else:\n",
        "        source_render_mask = Image.open(fetch(render_mask)).convert('RGBA')\n",
        "\n",
        "    if go_big_init_skipresize == False:\n",
        "\n",
        "      reside_x = side_x * go_big_scale\n",
        "      reside_y = side_y * go_big_scale\n",
        "      source_image = input_image.resize((reside_x, reside_y), Image.LANCZOS)\n",
        "      source_render_mask = source_render_mask.resize((reside_x, reside_y), Image.LANCZOS)\n",
        "    \n",
        "    else:\n",
        "      source_image = Image.open(fetch(progress_image)).convert('RGBA')\n",
        "\n",
        "    input_image.close()\n",
        "    # Slice source_image into overlapping slices\n",
        "    #print(f'debug: before slicing, slices_todo is {slices_todo}')\n",
        "    slices = slice(source_image, source_render_mask)\n",
        "    #print(f'debug: after slicing, slices_todo is {slices_todo}')\n",
        "    # Run PRD again for each slice, with init image paramaters, etc.\n",
        "    i = 1 # just to number the slices as they save\n",
        "    betterslices = []\n",
        "    seed_list =[]\n",
        "    init_seed = seed\n",
        "    for chunk, chunk_rmask in slices:\n",
        "        # Reset underlying systems for another run\n",
        "\n",
        "        seed = seed + 1\n",
        "        seed_list.append(seed)\n",
        "        args.seed = seed\n",
        "\n",
        "        # Go BIG Diffusion Model\n",
        "        print('Preparing model for next run...')\n",
        "        if go_big_diffusion_model != \"Inherit\":\n",
        "            print(go_big_diffusion_model)\n",
        "            dm_name = get_model_filename(go_big_diffusion_model)\n",
        "            print(\"Using model:\", dm_name)\n",
        "            model, diffusion = create_model_and_diffusion(**go_big_model_config)\n",
        "            model.load_state_dict(\n",
        "                torch.load(f'{model_path}/{dm_name}', map_location='cpu'))\n",
        "        else:\n",
        "            dm_name = get_model_filename(diffusion_model)\n",
        "            print(\"Using model:\", dm_name)\n",
        "            model, diffusion = create_model_and_diffusion(**model_config)\n",
        "            model.load_state_dict(\n",
        "                torch.load(f'{model_path}/{get_model_filename(diffusion_model)}', map_location='cpu'))\n",
        "        model.requires_grad_(False).eval().to(device)\n",
        "        for name, param in model.named_parameters():\n",
        "            if 'qkv' in name or 'norm' in name or 'proj' in name:\n",
        "                param.requires_grad_()\n",
        "        if model_config['use_fp16']:\n",
        "            model.convert_to_fp16()\n",
        "        gc.collect()\n",
        "        if \"cuda\" in str(device):\n",
        "          with torch.cuda.device(device):\n",
        "            torch.cuda.empty_cache()\n",
        "\n",
        "        #now do the next run\n",
        "        chunk.save(slice_image)\n",
        "        chunk_rmask.save(slice_rmask)\n",
        "        args.init_image = slice_image\n",
        "        args.skip_steps = int(steps * (gobig_skip_percent / 100))\n",
        "        args.side_x, args.side_y = chunk.size\n",
        "        args.symmetry_loss = False\n",
        "        args.render_mask = slice_rmask\n",
        "        render_mask = slice_rmask\n",
        "        do_run()\n",
        "        print(f'Finished run, grabbing {progress_image} and adding it to betterslices.')\n",
        "        resultslice = Image.open(progress_image).convert('RGBA')\n",
        "        betterslices.append(resultslice.copy())\n",
        "        resultslice.save(f'{batchFolder}/{batch_name}_slice_output_{current_time}_{i}.png')\n",
        "        resultslice.close() # hopefully this will allow subsequent images to actually save.\n",
        "        i += 1\n",
        "    # For each slice, use addalpha to add an alpha mask\n",
        "    alpha_gradient = Image.new('L', (args.side_x, 1), color=0xFF)\n",
        "    a = 0\n",
        "    for x in range(args.side_x):\n",
        "        a +=4 # add 4 to alpha at each pixel, to give us a 64 pixel overlap gradient\n",
        "        if a < 255:\n",
        "            alpha_gradient.putpixel((x, 0), a)\n",
        "        else:\n",
        "            alpha_gradient.putpixel((x, 0), 255)\n",
        "        # print '{}, {:.2f}, {}'.format(x, float(x) / width, a)\n",
        "    alpha = alpha_gradient.resize(betterslices[0].size, Image.BICUBIC)\n",
        "\n",
        "    mask = Image.new('RGBA', (args.side_x, args.side_y), color=0)\n",
        "    mask.putalpha(alpha)\n",
        "    i = 1 # start at 1 in the list instead of 0, because we don't need/want a mask on the first (0) image\n",
        "    while i < slices_todo:\n",
        "        betterslices[i] = addalpha(betterslices[i], mask)\n",
        "        i += 1\n",
        "    # Once we have all our images, mergeimgs back onto source.png, then save\n",
        "    final_output = mergeimgs(source_image, betterslices)\n",
        "    final_output.save(final_output_image)\n",
        "    print(f'\\n\\n\\033[32mGo BIG is Complete!\\33[0m\\n\\n\\33[33m\\u001b[4mNote:\\33[0m Your output is saved as {final_output_image}!')\n",
        "\n",
        "    \n",
        "  if go_big:\n",
        "    print(f\"\\nInitial Render Seed: {init_seed}\\nSlice Seeds:\")\n",
        "    print(*seed_list, sep = \", \") \n",
        "  else:\n",
        "    print('Seed used:', seed)\n",
        "\n",
        "except RuntimeError as e:\n",
        "\n",
        "  if 'out of memory' in str(e):\n",
        "    print(f\"\\u001b[31m\\u001b[1m\\u001b[4mCRITICAL ERROR\\u001b[0m: {gpu_name} ran out of memory! Flushing memory to save session...\")\n",
        "    if gpu_warning:\n",
        "      print(gpu_warning)\n",
        "    print('\\nSeed used for this failed run:', seed)\n",
        "    pass\n",
        "  else:\n",
        "    raise e\n",
        "\n",
        "except KeyboardInterrupt as e:\n",
        "\n",
        "  print('\\33[33mExecution interrupted by user.\\33[0m')\n",
        "  print('\\nSeed used for this exited run:', seed)\n",
        "  pass\n",
        "\n",
        "except Exception as e:\n",
        "\n",
        "  print(\"\\33[31mAn exxception as occured:\\33[0m\\n\", str(e))\n",
        "  if gpu_warning:\n",
        "    print(gpu_warning+\"\\n\")\n",
        "  print(\"Attempting to continue...\")\n",
        "  pass\n",
        "\n",
        "finally:\n",
        "\n",
        "    print(\"\\nCleaning up...\")\n",
        "    gc.collect()\n",
        "    torch.cuda.empty_cache()\n",
        "    print(\"Cleanup complete.\")"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CreateVidTop"
      },
      "source": [
        "# 6. Create the video"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CreateVid",
        "cellView": "form"
      },
      "source": [
        "import PIL\n",
        "# @title ### **Create video**\n",
        "#@markdown Video file will save in the same folder as your images.\n",
        "from tqdm.notebook import trange\n",
        "skip_video_for_run_all = False #@param {type: 'boolean'}\n",
        "\n",
        "if animation_mode == 'Video Input':\n",
        "    frames = sorted(glob(in_path+'/*.*'));\n",
        "    if len(frames)==0: \n",
        "        sys.exit(\"ERROR: 0 frames found.\\nPlease check your video input path and rerun the video settings cell.\")\n",
        "    flows = glob(flo_folder+'/*.*')\n",
        "    if (len(flows)==0) and video_init_flow_warp:\n",
        "        sys.exit(\"ERROR: 0 flow files found.\\nPlease rerun the flow generation cell.\")\n",
        "\n",
        "blend =  0.5#@param {type: 'number'}\n",
        "video_init_check_consistency = False #@param {type: 'boolean'}\n",
        "if skip_video_for_run_all == True:\n",
        "    print('Skipping video creation, uncheck skip_video_for_run_all if you want to run it')\n",
        "\n",
        "else:\n",
        "    # import subprocess in case this cell is run without the above cells\n",
        "    import subprocess\n",
        "    from base64 import b64encode\n",
        "\n",
        "    latest_run = batchNum\n",
        "\n",
        "    folder = batch_name #@param\n",
        "    run = latest_run #@param\n",
        "    final_frame = 'final_frame'\n",
        "\n",
        "\n",
        "    init_frame = 1#@param {type:\"number\"} This is the frame where the video will start\n",
        "    last_frame = final_frame#@param {type:\"number\"} You can change i to the number of the last frame you want to generate. It will raise an error if that number of frames does not exist.\n",
        "    fps = 12#@param {type:\"number\"}\n",
        "    # view_video_in_cell = True #@param {type: 'boolean'}\n",
        "\n",
        "    frames = []\n",
        "    # tqdm.write('Generating video...')\n",
        "\n",
        "    if last_frame == 'final_frame':\n",
        "        last_frame = len(glob(batchFolder+f\"/{folder}({run})_*.png\"))\n",
        "        print(f'Total frames: {last_frame}')\n",
        "\n",
        "    image_path = f\"{outDirPath}/{folder}/{folder}({run})_%04d.png\"\n",
        "    filepath = f\"{outDirPath}/{folder}/{folder}({run}).mp4\"\n",
        "\n",
        "    if (video_init_blend_mode == 'optical flow') and (animation_mode == 'Video Input'):\n",
        "        image_path = f\"{outDirPath}/{folder}/flow/{folder}({run})_%04d.png\"\n",
        "        filepath = f\"{outDirPath}/{folder}/{folder}({run})_flow.mp4\"\n",
        "        if last_frame == 'final_frame':\n",
        "            last_frame = len(glob(batchFolder+f\"/flow/{folder}({run})_*.png\"))\n",
        "        flo_out = batchFolder+f\"/flow\"\n",
        "        createPath(flo_out)\n",
        "        frames_in = sorted(glob(batchFolder+f\"/{folder}({run})_*.png\"))\n",
        "        shutil.copy(frames_in[0], flo_out)\n",
        "        for i in trange(init_frame, min(len(frames_in), last_frame)):\n",
        "            frame1_path = frames_in[i-1]\n",
        "            frame2_path = frames_in[i]\n",
        "  \n",
        "            frame1 = PIL.Image.open(frame1_path)\n",
        "            frame2 = PIL.Image.open(frame2_path)\n",
        "            frame1_stem = f\"{(int(frame1_path.split('/')[-1].split('_')[-1][:-4])+1):04}.jpg\"\n",
        "            flo_path = f\"/{flo_folder}/{frame1_stem}.npy\"\n",
        "            weights_path = None\n",
        "            if video_init_check_consistency:\n",
        "                # TBD\n",
        "                pass\n",
        "            warp(frame1, frame2, flo_path, blend=blend, weights_path=weights_path).save(batchFolder+f\"/flow/{folder}({run})_{i:04}.png\")\n",
        "    if video_init_blend_mode == 'linear':\n",
        "        image_path = f\"{outDirPath}/{folder}/blend/{folder}({run})_%04d.png\"\n",
        "        filepath = f\"{outDirPath}/{folder}/{folder}({run})_blend.mp4\"\n",
        "        if last_frame == 'final_frame':\n",
        "            last_frame = len(glob(batchFolder+f\"/blend/{folder}({run})_*.png\"))\n",
        "        blend_out = batchFolder+f\"/blend\"\n",
        "        createPath(blend_out)\n",
        "        frames_in = glob(batchFolder+f\"/{folder}({run})_*.png\")\n",
        "        shutil.copy(frames_in[0], blend_out)\n",
        "        for i in trange(1, len(frames_in)):\n",
        "            frame1_path = frames_in[i-1]\n",
        "            frame2_path = frames_in[i]\n",
        "    \n",
        "            frame1 = PIL.Image.open(frame1_path)\n",
        "            frame2 = PIL.Image.open(frame2_path)\n",
        "          \n",
        "            frame = PIL.Image.fromarray((np.array(frame1)*(1-blend) + np.array(frame2)*(blend)).astype('uint8')).save(batchFolder+f\"/blend/{folder}({run})_{i:04}.png\")\n",
        "\n",
        "\n",
        "    cmd = [\n",
        "        'ffmpeg',\n",
        "        '-y',\n",
        "        '-vcodec',\n",
        "        'png',\n",
        "        '-r',\n",
        "        str(fps),\n",
        "        '-start_number',\n",
        "        str(init_frame),\n",
        "        '-i',\n",
        "        image_path,\n",
        "        '-frames:v',\n",
        "        str(last_frame+1),\n",
        "        '-c:v',\n",
        "        'libx264',\n",
        "        '-vf',\n",
        "        f'fps={fps}',\n",
        "        '-pix_fmt',\n",
        "        'yuv420p',\n",
        "        '-crf',\n",
        "        '17',\n",
        "        '-preset',\n",
        "        'veryslow',\n",
        "        filepath\n",
        "    ]\n",
        "\n",
        "    process = subprocess.Popen(cmd, cwd=f'{batchFolder}', stdout=subprocess.PIPE, stderr=subprocess.PIPE)\n",
        "    stdout, stderr = process.communicate()\n",
        "    if process.returncode != 0:\n",
        "        print(stderr)\n",
        "        raise RuntimeError(stderr)\n",
        "    else:\n",
        "        print(\"The video is ready and saved to the images folder\")\n",
        "\n",
        "    # if view_video_in_cell:\n",
        "    #     mp4 = open(filepath,'rb').read()\n",
        "    #     data_url = \"data:video/mp4;base64,\" + b64encode(mp4).decode()\n",
        "    #     display.HTML(f'<video width=400 controls><source src=\"{data_url}\" type=\"video/mp4\"></video>')"
      ],
      "outputs": [],
      "execution_count": null
    }
  ],
  "metadata": {
    "anaconda-cloud": {},
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [
        "TutorialTop",
        "AnimSetTop",
        "ExtraSetTop"
      ],
      "machine_shape": "hm",
      "name": "DiscoStream Legacy",
      "private_outputs": true,
      "provenance": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.6.1"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}