Stable Diffusion API

Runs your own Stable Diffusion v1.4 API with Flask & Docker

Requirements:

• NVIDIA GPU with CUDA installed & >10GB memory
• Ubunutu 20.04 (this can be adjusted in Dockerfile if you are using a different OS)
• Huggingface Access Token

To Start the Server:

1. git clone https://github.com/viral-medialab/stable_diffusion_server.git
2. sh setup.sh
3. In line 34 of the Dockerfile, replace <TOKEN> with your Huggingface Access Token
4. docker build --tag sd_server .
5. docker run -d -p 3000:3000 sd_server The server should now be running at http://<your_external_ip>:3000/

To Generate an Image from Text:

• Pass your prompt as query string ('?prompt=') to the server
• When calling the server from a browser, be sure to replace spaces with '%20'
• Example Prompt: "Impressionist painting of dog wearing party hat."
  • Resulting url: http://<your_external_ip>:3000/?prompt=impressionist%20painting%20of%20dog%20wearing%20party%20hat
  • Wait ~30 sec for image generation and return
  • 

The foundation of this repo was built by following Lulia Turc's article.