Open Genmoji LoRAs

Downloading LoRAs

LoRA downloads are done through HuggingFace. Run the following Python script in the root open-genmoji directory to download a LoRA. The script will give you a list of available LoRAs, what models they use, and a brief description. Choose your favorite!

# If you're in the lora directory, cd out to root `open-genmoji` folder
cd ../

python3 ./download.py

LoRA List

List of metadata for all Genmoji LoRAs.

LoRA Name	Model Trained	Emojis Used	More Info
flux-dev.safetensors	Flux.1 Dev	No Skintone Apple Emoji List (1901)	View
diverse-emoji.safetensors	Flux.1 Dev	Complete Apple Emoji list (3770)	View

Contributing a LoRA

Contributing is easy! Here's the simple steps:

1. First, create a LoRA. You can learn how in README in the open-genmoji/finetuning folder in the root directory.

2. Go ahead and name that LoRA, and submit a PR to the Open Genmoji HuggingFace, adding that file.

3. If you would like your model to use a custom metaprompt for Prompt Assist, you should add it to the metaprompt folder. A metaprompt consists of a md file for an LLM to improve the prompt to the image model. You can also choose to include a LLM conversation history json. Both of these files should have the same file name (For example open-gemoji.md and open-genmoji.json). If you have a custom metaprompt md but no custom json, no Conversation History will be loaded. Note that there is already a metaprompt/open-genmoji JSON and Markdown. You may use that as your default.

   Optional: Improving Prompt Assistant consistency by adding a conversation history JSON

   To make outputs of your metaprompt more consistent across different LLM's, you can choose to add example combinations of input and output texts to a JSON file with the same name as your metaprompt.

   The file should be structured with a key "messages" which has a value containing a list of dictionaries as shown below.

   The content of the first message (with "role": "user") should include:

   • The full metaprompt
   • Two new lines \n\n
   • USER PROMPT:
   • The user prompt. This can be any non-optimized prompt.

   The content of the second message (with "role": "assistant") should include:

   • Only the optimized prompt.

   You can add multiple examples by repeating this. The file shown below has 2 example prompts and outputs.

      {
        "messages": [
          { "role": "user", "content": "You are helping create a prompt for a Emoji generation image model. An emoji must be easily interpreted when small so details must be exaggerated to be clear. Your goal is to use descriptions to achieve this.\n\nYou will receive a user description, and you must rephrase it to consist of short phrases separated by periods, adding detail to everything the user provides.\n\nAdd describe the color of all parts or components of the emoji. Unless otherwise specified by the user, do not describe people. Do not describe the background of the image. Your output should be in the format:\n\n```\nemoji of {description}. {addon phrases}. 3D lighting. no cast shadows.\n```\n\nThe description should be a 1 sentence of your interpretation of the emoji.\nThen, you may choose to add addon phrases. You must use the following in the given scenarios:\n\n- \"cute.\": If generating anything that's not an object, and also not a human\n- \"enlarged head in cartoon style.\": ONLY animals\n- \"head is turned towards viewer.\": ONLY humans or animals\n- \"detailed texture.\": ONLY objects\n\nFurther addon phrases may be added to ensure the clarity of the emoji.\n\n\n USER PROMPT: a horse wearing a suit"},
          { "role": "assistant", "content": "emoji of horse in black suit and tie with flowing mane. a strong, confident stallion wearing formal attire for a special occasion. cute. 3D lighting. no cast shadows. enlarged head in cartoon style. head is turned towards viewer." },
          { "role": "user", "content": "You are helping create a prompt for a Emoji generation image model. An emoji must be easily interpreted when small so details must be exaggerated to be clear. Your goal is to use descriptions to achieve this.\n\nYou will receive a user description, and you must rephrase it to consist of short phrases separated by periods, adding detail to everything the user provides.\n\nAdd describe the color of all parts or components of the emoji. Unless otherwise specified by the user, do not describe people. Do not describe the background of the image. Your output should be in the format:\n\n```\nemoji of {description}. {addon phrases}. 3D lighting. no cast shadows.\n```\n\nThe description should be a 1 sentence of your interpretation of the emoji.\nThen, you may choose to add addon phrases. You must use the following in the given scenarios:\n\n- \"cute.\": If generating anything that's not an object, and also not a human\n- \"enlarged head in cartoon style.\": ONLY animals\n- \"head is turned towards viewer.\": ONLY humans or animals\n- \"detailed texture.\": ONLY objects\n\nFurther addon phrases may be added to ensure the clarity of the emoji.\n\n\n USER PROMPT: flying pig"},
          { "role": "assistant", "content": "emoji of flying pink pig. enlarged head in cartoon style. cute. white wings. head is turned towards viewer. 3D lighting. no cast shadows." }
        ]
      }

   Why does this work?
   These messages are added to the conversation history with the chatbot. Then, when you submit a new prompt, it is likely to respond in a similar way it did before.

4. Next, you'll need to update info.json in this directory. Add an entry like such underneath all the other models:

{
	"model": "black-forest-labs/FLUX.1-dev", // HuggingFace path of base model
	"huggingface": "EvanZhouDev/open-genmoji", // Your own HuggingFace repo, optional
	"name": "flux-dev", // The name of your LoRA file
	"metaprompt": "open-genmoji", // The name of your metaprompt file (do not include .md or .json)
	"description": "Original Open Genmoji finetune. Trained on Apple Emojis w/o Skin Tone Variants." // Nice and short description of your model (< 100 characters)
}

5. Finally, update this file, to let people know what you trained on, how you trained it, etc. Start by adding a row with proper info in the LoRA List. Then, add a new section below in Detailed LoRA Description. The template for it can be found in LORA_TEMPLATE.md file in this directory.
6. With all that done, submit a PR to this repository. I'll get back to you as soon as possible.

Thank you for your contribution!

Detailed LoRA Description

flux-dev.safetensors

Model Finetuned: Flux.1 Dev Finished Finetuning: December 25, 2024 Finetuned By: Evan Zhou (@evanzhoudev) Finetuned On: M4 Max MacBook Pro (64GB Unified Memory, 40 core GPU) Finetuned With: SimpleTuner Finetuning Time: ~10 hours Finetuning Steps: 10,000 Image Resolution: 160x160 Emoji Dataset: All Apple Emojis as of December 25, 2024, with skin tone variants pruned to reduce training time.

diverse-emoji.safetensors

Model Finetuned: Flux.1 Dev Finished Finetuning: December 31, 2024 Finetuned By: Julius (@caspersimon) Finetuned On: M4 Max MacBook Pro (128GB Unified Memory, 40 core GPU) Finetuned With: SimpleTuner Finetuning Time: ~20 hours Finetuning Steps: 20,000 Image Resolution: 160x160 Emoji Dataset: All Apple Emoji as of December 30, 2024. Huggingface Link: caspersimon/Diverse-Emoji