Render live
Strokes go up and frames come down over one WebSocket.
Draw it.
potocolom turns your canvas into finished images while you sketch. Run it on your own GPU for free, or let the cloud carry the weight.
One canvas, four jobs
Generation, upscaling, editing and the live loop run through the same job path, the same models and the same history, self-hosted or in the cloud.
Generate
Text to image. Pick a model and the controls come from its manifest, so a new model needs no frontend release.
Upscale
Take a finished image up in size with Real-ESRGAN class weights. Fast tier by default, quality tier when it matters.
Edit
Image to image: feed a picture back in with a prompt and a strength, and keep the parts that already work.
What people are making
Sketches become finished images in real time. A sample of work from the community - the kind of output the canvas produces.
Hover to pause. Click any prompt to copy it.
Simple tiers, honest credits
Credits meter GPU seconds transparently: drawing drains them live, generations cost a handful each. Beta pricing, subject to change before public launch - the waitlist hears first.
Starter
€9 per month
- 10,000 credits every month
- About 2.7 hours of live drawing
- Or a few thousand quick generations
Pro Most popular
€24 per month
- 35,000 credits every month
- About 9.5 hours of live drawing
- Priority in the queue
Studio
€59 per month
- 100,000 credits every month
- About 27 hours of live drawing
- Top queue priority
Enterprise
Talk to us
- Dedicated fleet capacity
- Annual invoicing and volume credits
- Support for your self-hosted install
Every beta invite includes a one-time trial grant - no card needed to try.
Take the repo. Run it on your GPU.
- The realtime drawing loop is in the repo: strokes up, renders down, on your hardware.
- New models arrive as manifests. The interface reads their schema - no waiting on a release.
- One compose file on NVIDIA or AMD. No account required, no public gallery, no phone-home.
zsh - ~/potocolom
~/potocolom $ █
A few things to know
The short version of how the studio, the hardware and the hosted service fit together.
What is potocolom, exactly?
An open source image studio built around a live drawing loop. You sketch and write a prompt; a diffusion model continuously re-renders the canvas. The same studio also handles generation, editing and upscaling.
Is self-hosting a limited version?
No. The AGPL 3.0 repository contains the complete application. Run it with Docker Compose on a supported NVIDIA or AMD GPU, without an account or required phone-home. The paid cloud is the same product operated for you.
Which models and GPUs can I use?
Models arrive through manifests that declare their controls, licenses and VRAM needs. The current Studio catalogue includes SDXL-based Stability AI models, realtime and fast variants, and Real-ESRGAN upscalers on NVIDIA or AMD hardware. The benchmark page publishes measured requirements and speeds.
What happens to my images?
Assets are private by default and served through short-lived signed links. Sharing is explicit, your data can be exported, and your work is not used to train models.
How will the cloud beta and credits work?
The cloud meters actual GPU time. Active drawing and GPU work consume credits; an idle canvas does not, and failed work refunds automatically. Beta invitations include a trial grant with no card required.
Be first on the canvas
The cloud opens as an invite-only beta. Join the waitlist and you will hear exactly twice: at the self-host launch and when your invite is ready.
One address, one purpose, deletable on request.