vacation/gonk-diffusers-api repository overview

⁠gonk-diffusers-api

An API server to transform text into images using using the diffusers library on CUDA hardware.

⁠Example request in a Jupyter notebook

import requests
import base64
from IPython.display import Image, display

url = 'http://localhost:8000/generate-image'
data = {
    'model': 'dataautogpt3/OpenDalleV1.1',
    'prompt': 'Aerial view of a futuristic cityscape getting tip-toed on by a giant kitten acting like godzilla',
    'negative_prompt': "(worst quality, low quality, simpsons hands)",
    'width': 1024,
    'height': 1024,
    'image_type': 'png',
    'num_inference_steps': 40,
    'safety': True
}
response = requests.post(url, json=data)
response_data = response.json()

# Check the response and display the image
if response.status_code == 200 and 'image' in response_data:
    encoded_image = response_data['image']
    mime_type = response_data['mime_type']
    
    # Decode the Base64 string to bytes
    image_bytes = base64.b64decode(encoded_image)
    
    # Display the image
    display(Image(data=image_bytes, format=mime_type.split('/')[-1], embed=True))
else:
    print("Failed to generate image:", response.text)

⁠Running the Docker

Build the docker image

$ docker build --no-cache -t gonk-diffusers-api .

Then run the container, be sure nvidia docker runtime is installed.

$ docker run --runtime=nvidia \
             --rm -p 8000:8000 \
             -v /path/to/huggingface-hub-cache:/path/to/huggingface-hub-cache \
             gonk-diffusers-api \
             --hf-cache-path /path/to/huggingface-hub-cache \
             --hf-local-files-only

Note: The huggingface-hub-cache refers to the directory holding the models--.. folders.

⁠Features

• Stable Diffusion Compatibility: Utilizes models compatible with Stable Diffusion for high-quality image generation.
• CUDA GPU Acceleration: Optimized for use with CUDA-compatible NVIDIA GPUs.
• On-by-default Safety: Leverages the Falconsai/nsfw_image_detection⁠ model to blur images that do not pass the safety filter.
• Concurrency Control: Manages multiple simultaneous requests with built-in concurrency control.
• Interactive Swagger Documentation: Includes Swagger UI for easy interaction with the API.

⁠System Requirements

• CUDA-Compatible GPU: Requires an NVIDIA GPU with CUDA support.
• Python 3.8+: Recommended to use the latest version of Python.
• Dependencies: Listed in the requirements.txt file.

⁠Installation

1. Clone the Repository:

   git clone https://github.com/jason-weirather/gonk-diffusers-api.git
   

   Copy
2. Navigate to the Project Directory:

   cd gonk-diffusers-api
   

   Copy
3. Install Dependencies:

   pip install -r requirements.txt
   

   Copy

⁠Usage

⁠Starting the Server

Start the server using the CLI:

python -m gonk-diffusers-api.cli

Optional CLI arguments:

• -H, --host: Host address (default 0.0.0.0).
• -p, --port: Port number (default 8000).
• --require-auth: Enable API key authentication.

⁠API Endpoints

• POST /generate-image: Generates an image from a given text prompt.
• GET /status: Retrieves the server and CUDA status.

⁠Authentication

To enable API authentication, set the GONK_DIFFUSERS_API_KEY environment variable and start the server with the --require-auth flag.

⁠API Documentation

Access the interactive API documentation by navigating to http://localhost:8000/docs.

⁠License

This project is licensed under the Apache License 2.0⁠.