michalautumn8/stable-diffusion-fp16-gpu-only repository overview

https://huggingface.co/stabilityai/stable-diffusion-2⁠

⁠Usage:

⁠Prerequisites

needs to be run on a machine with:

• an NVIDIA GPU available,
• CUDA drivers installed,
• nvidia-container-toolkit installed

For testing, I used an AWS EC2 g5.2xlarge, configured with AMI

and installed Docker + nvidia-container-toolkit by following this guide (click me, I'm a link)⁠

⁠Startup:

docker run --gpus all -p 4567:8080 michalautumn8/stable-diffusion-fp16-gpu-only

⁠Request:

curl -X POST http://localhost:4567/v1/models/autodl:predict \
        -H "Content-Type: application/json" \
        -d '"a photo of an astronaut riding a horse on mars"'

⁠Response:

Returns a 1MB base64-encoded string, which, after conversion - in our example - corresponds to the following image:

⁠Shortcomings

• does not support batch IO
  • that should be simple to fix by editing the handler file included in the image