Austin French
← Playground

Image generation

Text to image with Stable Diffusion 1.5, run through ComfyUI.

This tool ran on a GPU at my place, and it's been offline since I moved to New York. Below is how it worked. The engineering behind all of them is in austn.net's GPU tools.

model
Stable Diffusion 1.5 (fp16) on ComfyUI

Type a prompt, pick a size, get an image.

How it worked

  1. The page posts the prompt to Rails, which validates and clamps the size and batch settings and queues a job on the gpu queue.
  2. The job waits for the GPU lock, fills in a saved ComfyUI workflow with the prompt and settings, and submits it.
  3. It polls ComfyUI until the image is ready, stores the result in Redis for a short time, and marks the request complete.
  4. The page polls the status endpoint and shows the image when it's done.

A default negative prompt keeps out the usual artifacts: blur, pixelation, compression noise.