feedback-image-trainer 0.1.0

Fine-tune Stable Diffusion with feedback-driven Optuna hyperparameter search

overview

# Feedback Image Trainer

A Python package to fine-tune Stable Diffusion models using feedback-driven hyperparameter search with Optuna.

## Installation

```bash
pip install feedback_image_trainer

Usage

from feedback_image_trainer import run_study

run_study(
    feedback_file="image_feedback.json",
    model_path="runwayml/stable-diffusion-v1-5",
    output_dir="fine_tuned_model",
    trials=5
)

Input Data Format

The image_feedback.json file should contain a list of dictionaries with the following structure:

[
    {
        "image_path": "path/to/image.png",
        "prompt": "A description of the image",
        "feedback": 1
    },
    ...
]

Requirements

• Python 3.8+
• torch>=2.0.0
• diffusers>=0.20.0
• transformers>=4.30.0
• accelerate>=0.20.0
• optuna>=2.0.0
• torchvision>=0.15.0