fing

A REST API for stable diffusion

(though I've stopped working on this and now spend more time over here

Introduction

This is a REST interface to stable-diffusion.

What you will need:

• A machine with a decent CUDA capable graphics card. (NVIDIA GTX 1080, RTX 3060 or so - 8GB or more)
• Anaconda version 2022.05 (might work on older versions) on that machine
• I've done this all on ubuntu 22.04. Should work on other linuxes.
• A Huggingface account and access token
  • You will also need to accept the model license agreement

Features

• Includes txt2img, img2img, imginpaint, and face generation
• Pipeline cache so any of the above can be swapped into the GPU as needed
• GPU pooling to support multiple concurrent requests (1 per GPU)
• Optional x-api-key for simple auth
• Open API swagger spec
  • once the server is running the latest spec can always be viewed at <server_address:port>/docs

Setup

cd src
conda env create -f environment.yaml
conda activate fing
python -m generator.init_app

This will create settings.json file in ~/.fing as well as download all of the diffuser model caches. This last part can take some time.

Running the Service

The rest service can be started this way:

cd src
conda activate fing

python -m generator.server
or 
uvicorn generator.server:app --host 0.0.0.0 --port 9147

This will run a simple REST api on port 9147:

<div> http://localhost:3010/api/txt2img?prompt=Proof of space and time </div> <br> <img src="post.jpg" width="256" height="256" alt="Proof of space and time."/>