Diffusers FastAPI

Diffusers FastAPI is a robust and versatile FastAPI-based server for image generation, image-to-image transformation, and inpainting using Hugging Face Diffusers. This project provides a powerful API for various image generation and manipulation tasks using state-of-the-art diffusion models.

Features

• Text-to-image generation
• Image-to-image transformation
• Inpainting
• Support for multiple Stable Diffusion models
• Automatic package management and updates
• Configurable output and model caching directories
• Verbose logging option for debugging

Installation

1. Clone the repository:

   git clone https://github.com/ParisNeo/diffusers-fastapi.git
   cd diffusers-fastapi
   

2. Install PyTorch: This application requires PyTorch to be installed, preferably with CUDA support for optimal performance. To install PyTorch with CUDA support, use the following command:

   pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
   

   This command installs PyTorch with CUDA 11.8 support. If you need a different CUDA version or don't have a CUDA-capable GPU, please refer to the official PyTorch installation guide for alternative installation options.

   Note: The CUDA version (cu118 in this case) should match your system's CUDA installation. Adjust as necessary.

3. Install other dependencies:

   pip install -r requirements.txt
   

Usage

To start the Diffusers FastAPI server, run the following command:

python diffusers_fastapi.py [options]

Options

• --host: Host to run the server on (default: 127.0.0.1)
• --port: Port to run the server on (default: 8253)
• --model: Diffusers model to use (default: "v2ray/stable-diffusion-3-medium-diffusers")
• --output_dir: Directory to save generated images (default: "output")
• --models_dir: Directory to cache models (default: "models")
• --verbose: Enable verbose logging

Example:

python diffusers_fastapi.py --host 0.0.0.0 --port 8253 --model "stabilityai/stable-diffusion-2-1" --output_dir "output" --models_dir "models" --verbose

API Endpoints

The server provides the following endpoints:

1. /generate-image: Text-to-image generation
2. /img2img: Image-to-image transformation
3. /inpaint: Inpainting

For detailed information about request and response formats, please refer to the API documentation available at /docs when the server is running.

Python Examples

Here are some examples of how to use the API with Python requests:

Text-to-Image Generation

import requests
import json

url = "http://localhost:8253/generate-image"
payload = {
    "positive_prompt": "A beautiful landscape with mountains and a lake",
    "negative_prompt": "clouds, people",
    "seed": 42,
    "scale": 7.5,
    "steps": 20,
    "width": 512,
    "height": 512
}

response = requests.post(url, json=payload)
result = json.loads(response.text)
print(f"Generated image path: {result['image_path']}")

Image-to-Image Transformation

import requests

url = "http://localhost:8253/img2img"
files = {
    'image': ('input.png', open('input.png', 'rb'), 'image/png')
}
data = {
    'positive_prompt': 'Transform this landscape into a snowy scene',
    'negative_prompt': 'summer, green',
    'seed': 42,
    'scale': 7.5,
    'steps': 20
}

response = requests.post(url, files=files, data=data)
result = response.json()
print(f"Generated image path: {result['image_path']}")

Inpainting

import requests

url = "http://localhost:8253/inpaint"
files = {
    'image': ('input.png', open('input.png', 'rb'), 'image/png'),
    'mask': ('mask.png', open('mask.png', 'rb'), 'image/png')
}
data = {
    'positive_prompt': 'Add a cat sitting on the couch',
    'negative_prompt': 'dog, bird',
    'seed': 42,
    'scale': 7.5,
    'steps': 20
}

response = requests.post(url, files=files, data=data)
result = response.json()
print(f"Generated image path: {result['image_path']}")

API Documentation

For detailed API documentation, including request and response schemas, please run the server and navigate to http://localhost:8253/docs in your web browser. This will open the Swagger UI, which provides interactive documentation for all available endpoints.

License

This project is licensed under the Apache 2.0 License.

Author

ParisNeo - A computer geek passionate about AI

GitHub: https://github.com/ParisNeo

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.