predict_t2v.py

import os

import numpy as np
import torch
from diffusers import (DDIMScheduler, DPMSolverMultistepScheduler,
                       EulerAncestralDiscreteScheduler, EulerDiscreteScheduler,
                       FlowMatchEulerDiscreteScheduler, PNDMScheduler)
from omegaconf import OmegaConf
from PIL import Image
from transformers import (BertModel, BertTokenizer, CLIPImageProcessor,
                          CLIPVisionModelWithProjection, Qwen2Tokenizer,
                          Qwen2VLForConditionalGeneration, T5EncoderModel,
                          T5Tokenizer)

from easyanimate.models import (name_to_autoencoder_magvit,
                                name_to_transformer3d)
from easyanimate.models.transformer3d import get_teacache_coefficients
from easyanimate.pipeline.pipeline_easyanimate import EasyAnimatePipeline
from easyanimate.pipeline.pipeline_easyanimate_inpaint import \
    EasyAnimateInpaintPipeline
from easyanimate.utils.fp8_optimization import (convert_model_weight_to_float8,
                                                convert_weight_dtype_wrapper)
from easyanimate.utils.lora_utils import merge_lora, unmerge_lora
from easyanimate.utils.utils import get_image_to_video_latent, save_videos_grid

# GPU memory mode, which can be choosen in [model_cpu_offload, model_cpu_offload_and_qfloat8, sequential_cpu_offload].
# model_cpu_offload means that the entire model will be moved to the CPU after use, which can save some GPU memory.
# 
# model_cpu_offload_and_qfloat8 indicates that the entire model will be moved to the CPU after use, 
# and the transformer model has been quantized to float8, which can save more GPU memory. 
# 
# sequential_cpu_offload means that each layer of the model will be moved to the CPU after use, 
# resulting in slower speeds but saving a large amount of GPU memory.
# 
# EasyAnimateV1, V2 and V3 support "model_cpu_offload" "sequential_cpu_offload"
# EasyAnimateV4, V5 and V5.1 support "model_cpu_offload" "model_cpu_offload_and_qfloat8" "sequential_cpu_offload"
GPU_memory_mode     = "model_cpu_offload_and_qfloat8"
# EasyAnimateV5.1 support TeaCache.
enable_teacache     = True
# Recommended to be set between 0.05 and 0.1. A larger threshold can cache more steps, speeding up the inference process, 
# but it may cause slight differences between the generated content and the original content.
teacache_threshold  = 0.08

# Config and model path
config_path         = "config/easyanimate_video_v5.1_magvit_qwen.yaml"
model_name          = "models/Diffusion_Transformer/EasyAnimateV5.1-12b-zh-InP"

# Choose the sampler in "Euler" "Euler A" "DPM++" "PNDM" "DDIM" "Flow"
# EasyAnimateV1, V2 and V3 support "Euler" "Euler A" "DPM++" "PNDM"
# EasyAnimateV4 and V5 support "Euler" "Euler A" "DPM++" "PNDM" "DDIM".
# EasyAnimateV5.1 supports Flow.
sampler_name        = "Flow"

# Load pretrained model if need
transformer_path    = None
# Only V1 does need a motion module
motion_module_path  = None
vae_path            = None
lora_path           = None

# Other params
sample_size         = [384, 672]
# In EasyAnimateV1, the video_length of video is 40 ~ 80.
# In EasyAnimateV2, V3, V4, the video_length of video is 1 ~ 144. 
# In EasyAnimateV5, V5.1, the video_length of video is 1 ~ 49.
# If u want to generate a image, please set the video_length = 1.
video_length        = 49
fps                 = 8

# Use torch.float16 if GPU does not support torch.bfloat16
# ome graphics cards, such as v100, 2080ti, do not support torch.bfloat16
weight_dtype        = torch.bfloat16

# EasyAnimateV1, V2 and V3 support English.
# EasyAnimateV4, V5 and V5.1 support English and Chinese.
# 使用更长的neg prompt如"模糊，突变，变形，失真，画面暗，文本字幕，画面固定，连环画，漫画，线稿，没有主体。"，可以增加稳定性
# 在neg prompt中添加"安静，固定"等词语可以增加动态性。
prompt              = "一只棕褐色的狗正摇晃着脑袋，坐在一个舒适的房间里的浅色沙发上。沙发看起来柔软而宽敞，为这只活泼的狗狗提供了一个完美的休息地点。在狗的后面，靠墙摆放着一个架子，架子上挂着一幅精美的镶框画，画中描绘着一些美丽的风景或场景。画框周围装饰着粉红色的花朵，这些花朵不仅增添了房间的色彩，还带来了一丝自然和生机。房间里的灯光柔和而温暖，从天花板上的吊灯和角落里的台灯散发出来，营造出一种温馨舒适的氛围。整个空间给人一种宁静和谐的感觉，仿佛时间在这里变得缓慢而美好。"
negative_prompt     = "扭曲的身体，肢体残缺，文本字幕，漫画，静止，丑陋，错误，乱码。"
# 
# Using longer neg prompt such as "Blurring, mutation, deformation, distortion, dark and solid, comics, text subtitles, line art." can increase stability
# Adding words such as "quiet, solid" to the neg prompt can increase dynamism.
# prompt              = "The dog is shaking head. The video is of high quality, and the view is very clear. High quality, masterpiece, best quality, highres, ultra-detailed, fantastic."
# negative_prompt     = "Twisted body, limb deformities, text captions, comic, static, ugly, error, messy code."
guidance_scale      = 6.0
seed                = 43
num_inference_steps = 50
lora_weight         = 0.60
save_path           = "samples/easyanimate-videos"

config = OmegaConf.load(config_path)

# Get Transformer
Choosen_Transformer3DModel = name_to_transformer3d[
    config['transformer_additional_kwargs'].get('transformer_type', 'Transformer3DModel')
]

transformer_additional_kwargs = OmegaConf.to_container(config['transformer_additional_kwargs'])
if weight_dtype == torch.float16 and "v5.1" not in model_name.lower():
    transformer_additional_kwargs["upcast_attention"] = True

transformer = Choosen_Transformer3DModel.from_pretrained_2d(
    model_name, 
    subfolder="transformer",
    transformer_additional_kwargs=transformer_additional_kwargs,
    torch_dtype=torch.float8_e4m3fn if GPU_memory_mode == "model_cpu_offload_and_qfloat8" else weight_dtype,
    low_cpu_mem_usage=True,
)

if transformer_path is not None:
    print(f"From checkpoint: {transformer_path}")
    if transformer_path.endswith("safetensors"):
        from safetensors.torch import load_file, safe_open
        state_dict = load_file(transformer_path)
    else:
        state_dict = torch.load(transformer_path, map_location="cpu")
    state_dict = state_dict["state_dict"] if "state_dict" in state_dict else state_dict

    m, u = transformer.load_state_dict(state_dict, strict=False)
    print(f"missing keys: {len(m)}, unexpected keys: {len(u)}")

if motion_module_path is not None:
    print(f"From Motion Module: {motion_module_path}")
    if motion_module_path.endswith("safetensors"):
        from safetensors.torch import load_file, safe_open
        state_dict = load_file(motion_module_path)
    else:
        state_dict = torch.load(motion_module_path, map_location="cpu")
    state_dict = state_dict["state_dict"] if "state_dict" in state_dict else state_dict

    m, u = transformer.load_state_dict(state_dict, strict=False)
    print(f"missing keys: {len(m)}, unexpected keys: {len(u)}, {u}")

# Get Vae
Choosen_AutoencoderKL = name_to_autoencoder_magvit[
    config['vae_kwargs'].get('vae_type', 'AutoencoderKL')
]
vae = Choosen_AutoencoderKL.from_pretrained(
    model_name, 
    subfolder="vae",
    vae_additional_kwargs=OmegaConf.to_container(config['vae_kwargs'])
).to(weight_dtype)
if weight_dtype == torch.float16 and "v5.1" not in model_name.lower():
    vae.upcast_vae = True

if vae_path is not None:
    print(f"From checkpoint: {vae_path}")
    if vae_path.endswith("safetensors"):
        from safetensors.torch import load_file, safe_open
        state_dict = load_file(vae_path)
    else:
        state_dict = torch.load(vae_path, map_location="cpu")
    state_dict = state_dict["state_dict"] if "state_dict" in state_dict else state_dict

    m, u = vae.load_state_dict(state_dict, strict=False)
    print(f"missing keys: {len(m)}, unexpected keys: {len(u)}")

if config['text_encoder_kwargs'].get('enable_multi_text_encoder', False):
    tokenizer = BertTokenizer.from_pretrained(
        model_name, subfolder="tokenizer"
    )
    if config['text_encoder_kwargs'].get('replace_t5_to_llm', False):
        tokenizer_2 = Qwen2Tokenizer.from_pretrained(
            os.path.join(model_name, "tokenizer_2")
        )
    else:
        tokenizer_2 = T5Tokenizer.from_pretrained(
            model_name, subfolder="tokenizer_2"
        )
else:
    if config['text_encoder_kwargs'].get('replace_t5_to_llm', False):
        print(os.path.join(model_name, "tokenizer"))
        tokenizer = Qwen2Tokenizer.from_pretrained(
            os.path.join(model_name, "tokenizer")
        )
    else:
        tokenizer = T5Tokenizer.from_pretrained(
            model_name, subfolder="tokenizer"
        )
    tokenizer_2 = None

if config['text_encoder_kwargs'].get('enable_multi_text_encoder', False):
    text_encoder = BertModel.from_pretrained(
        model_name, subfolder="text_encoder"
    ).to(weight_dtype)
    if config['text_encoder_kwargs'].get('replace_t5_to_llm', False):
        text_encoder_2 = Qwen2VLForConditionalGeneration.from_pretrained(
            os.path.join(model_name, "text_encoder_2"),
            torch_dtype=weight_dtype,
        )
    else:
        text_encoder_2 = T5EncoderModel.from_pretrained(
            model_name, subfolder="text_encoder_2"
        ).to(weight_dtype)
else:
    if config['text_encoder_kwargs'].get('replace_t5_to_llm', False):
        text_encoder = Qwen2VLForConditionalGeneration.from_pretrained(
            os.path.join(model_name, "text_encoder"),
            torch_dtype=weight_dtype,
        )
    else:
        text_encoder = T5EncoderModel.from_pretrained(
            model_name, subfolder="text_encoder"
        ).to(weight_dtype)
    text_encoder_2 = None

if transformer.config.in_channels != vae.config.latent_channels and config['transformer_additional_kwargs'].get('enable_clip_in_inpaint', True):
    clip_image_encoder = CLIPVisionModelWithProjection.from_pretrained(
        model_name, subfolder="image_encoder"
    ).to(weight_dtype)
    clip_image_processor = CLIPImageProcessor.from_pretrained(
        model_name, subfolder="image_encoder"
    )
else:
    clip_image_encoder = None
    clip_image_processor = None

# Get Scheduler
Choosen_Scheduler = scheduler_dict = {
    "Euler": EulerDiscreteScheduler,
    "Euler A": EulerAncestralDiscreteScheduler,
    "DPM++": DPMSolverMultistepScheduler, 
    "PNDM": PNDMScheduler,
    "DDIM": DDIMScheduler,
    "Flow": FlowMatchEulerDiscreteScheduler,
}[sampler_name]

scheduler = Choosen_Scheduler.from_pretrained(
    model_name, 
    subfolder="scheduler"
)

if transformer.config.in_channels != vae.config.latent_channels:
    pipeline = EasyAnimateInpaintPipeline(
        text_encoder=text_encoder,
        text_encoder_2=text_encoder_2,
        tokenizer=tokenizer,
        tokenizer_2=tokenizer_2,
        vae=vae,
        transformer=transformer,
        scheduler=scheduler,
        clip_image_encoder=clip_image_encoder,
        clip_image_processor=clip_image_processor,
    )
else:
    pipeline = EasyAnimatePipeline(
        text_encoder=text_encoder,
        text_encoder_2=text_encoder_2,
        tokenizer=tokenizer,
        tokenizer_2=tokenizer_2,
        vae=vae,
        transformer=transformer,
        scheduler=scheduler,
    )

if GPU_memory_mode == "sequential_cpu_offload":
    pipeline._manual_cpu_offload_in_sequential_cpu_offload = []
    for name, _text_encoder in zip(["text_encoder", "text_encoder_2"], [pipeline.text_encoder, pipeline.text_encoder_2]):
        if isinstance(_text_encoder, Qwen2VLForConditionalGeneration):
            if hasattr(_text_encoder, "visual"):
                del _text_encoder.visual
            convert_model_weight_to_float8(_text_encoder)
            convert_weight_dtype_wrapper(_text_encoder, weight_dtype)
            pipeline._manual_cpu_offload_in_sequential_cpu_offload = [name]
    pipeline.enable_sequential_cpu_offload()
elif GPU_memory_mode == "model_cpu_offload_and_qfloat8":
    for _text_encoder in [pipeline.text_encoder, pipeline.text_encoder_2]:
        if hasattr(_text_encoder, "visual"):
            del _text_encoder.visual
    convert_weight_dtype_wrapper(transformer, weight_dtype)
    pipeline.enable_model_cpu_offload()
else:
    pipeline.enable_model_cpu_offload()

coefficients = get_teacache_coefficients(model_name)
if coefficients is not None and enable_teacache:
    print(f"Enable TeaCache with threshold: {teacache_threshold}.")
    pipeline.transformer.enable_teacache(num_inference_steps, teacache_threshold, coefficients=coefficients)

generator = torch.Generator(device="cuda").manual_seed(seed)

if lora_path is not None:
    pipeline = merge_lora(pipeline, lora_path, lora_weight, device="cuda", dtype=weight_dtype)

with torch.no_grad():
    if transformer.config.in_channels != vae.config.latent_channels:
        if vae.cache_mag_vae:
            video_length = int((video_length - 1) // vae.mini_batch_encoder * vae.mini_batch_encoder) + 1 if video_length != 1 else 1
        else:
            video_length = int(video_length // vae.mini_batch_encoder * vae.mini_batch_encoder) if video_length != 1 else 1
        input_video, input_video_mask, clip_image = get_image_to_video_latent(None, None, video_length=video_length, sample_size=sample_size)

        sample = pipeline(
            prompt, 
            video_length = video_length,
            negative_prompt = negative_prompt,
            height      = sample_size[0],
            width       = sample_size[1],
            generator   = generator,
            guidance_scale = guidance_scale,
            num_inference_steps = num_inference_steps,

            video        = input_video,
            mask_video   = input_video_mask,
            clip_image   = clip_image, 
        ).frames
    else:
        sample = pipeline(
            prompt, 
            video_length = video_length,
            negative_prompt = negative_prompt,
            height      = sample_size[0],
            width       = sample_size[1],
            generator   = generator,
            guidance_scale = guidance_scale,
            num_inference_steps = num_inference_steps,
        ).frames

if lora_path is not None:
    pipeline = unmerge_lora(pipeline, lora_path, lora_weight, device="cuda", dtype=weight_dtype)

if not os.path.exists(save_path):
    os.makedirs(save_path, exist_ok=True)

index = len([path for path in os.listdir(save_path)]) + 1
prefix = str(index).zfill(8)

if video_length == 1:
    video_path = os.path.join(save_path, prefix + ".png")

    image = sample[0, :, 0]
    image = image.transpose(0, 1).transpose(1, 2)
    image = (image * 255).numpy().astype(np.uint8)
    image = Image.fromarray(image)
    image.save(video_path)
else:
    video_path = os.path.join(save_path, prefix + ".mp4")
    save_videos_grid(sample, video_path, fps=fps)