Spool
A workbench for making AI images and video on my own PC. A web page queues the jobs, a ComfyUI back end runs them one at a time, and everything stays on the machine.
Why
I wanted image and video generation without subscriptions, uploads or someone else's content rules, and with a queue so I can line up work and walk away. ComfyUI does the heavy lifting but is a node editor; Spool puts a plain form in front of it for the handful of jobs I actually run.
What it does
- Generate: text to image, with several versions per run on consecutive seeds, across a few engines chosen per job.
- Edit: change a picture with an instruction, restore or upscale it, relight it, or make a new picture in the style of up to three reference images. Poses and depth maps can be extracted from a picture and fed back in as references.
- Video: short clips with sound, from a prompt alone, a start picture, or a set of reference pictures. A clip can continue from where the last one ended.
- Jobs and prompts: a queue with a live status bar, a job list, and a prompt book that remembers and names what worked.
- Claude Code can use it. A skill submits jobs to the same queue over the same HTTP API, so an assistant can ask for icons, cover art or a contact sheet while I use the page myself.
Under the hood
- A small Python server with a plain web page. It starts ComfyUI in the background when the first job arrives and talks to it over its API.
- One job runs at a time. The models stay in GPU memory between jobs so the next one is quick, and an Unload button frees the card when I need it for something else.
- Runs on a single consumer GPU. Large models are used in compressed forms and the editing and video engines are swapped in and out as needed.
- The unit tests run with no models loaded and no server started, so they are fast and safe to run any time.
A personal tool that runs on my own PC and is not available to use. The screenshot shows images made for the purpose.