Raibows / WebPlayground-of-StableDiffusion

An Easy Deployment Web Playground of Stable Diffusion Model

An Easy Deployment Web Playground of Stable Diffusion Model

This repo provides a web playground which can be easily deployed of stable diffusion model.

Stable diffusion model is a state-of-the-art method for generating images from texts.

There are some platforms (e.g., Huggingface) providing such service for users to play online. However, they often do not provide full custom options for users and thus these models cannot be freely used.

This repo based on Huggingface web demo provides an easy way to deploy your playground and unlock all options for the diffusion model.

Requirements

1. Git clone this repo.

2. You need to obtain the auth token from Huggingface to access the pre-trained diffusion model. Please refer here.

3. Ubuntu with cuda supported.

4. Tested on the following python environment:

   Python==3.8
   cudatoolkit==11.1.1
   diffusers==0.4.1
   fastapi==0.85.0
   gradio @ https://gradio-builds.s3.amazonaws.com/queue-disconnect/v3/gradio-3.4b2-py3-none-any.whl
   huggingface-hub==0.10.0
   Pillow==9.2.0
   torch==1.9.0
   transformers==4.22.2
   

   Please install them using conda or pip.

Launch your playground

Launch it using:

python app.py --mode full --device 0 --auth your_auth_token --port 7890 --host 127.0.0.1

Open your browser with 127.0.0.1:7890, and have fun.

Full usage:

python app.py -h

usage: app.py [-h] [--mode {fp16,full,test}] [--host HOST] [--port PORT] [--device DEVICE] [--auth AUTH]

optional arguments:
  -h, --help              show this help message and exit
  --mode {fp16,full,test} test mode will not load the weights; fp16 loads a revised model that help accelerate the process but may be fixed to the image size 512x512; full indicates the original fp32 weights, more precise but more cost.
  --host HOST             specify the ip address
  --port PORT             specify the port
  --device DEVICE         if you have multiple devices, specify it as 0, 1, etc.
  --auth AUTH             fill it with yours in huggingface account to download diffusion model weights

If you have any questions, please feel free to have an issue or discussion.

Reference

Huggingface/stable-diffusion

Reddit/tutorial-to-unlock-stable-diffusion

License

The codes for web playground is under GPLv3 license.

The license for stable diffusion model, please refer this.