Possibility Studio. Natural-language creative coding Source

Possibility Studio / six screens

Describe it.
Pick a direction.

Describe the thing in plain English. You get three working HTML prototypes, each with a different layout and motion idea. Pick one, describe the changes you want, and export the file. Runs locally against any OpenAI-compatible provider.

01Brief

Describe it.

One paragraph in plain English: what it is, who it is for, and how it should feel. Add a reference image if you have one.

Possibility Studio: briefFull size
The brief form and recent projects.
02Directions

Pick your path.

Three directions come back, each with a different layout and motion idea. Read the tradeoff, then choose the one to build.

Possibility Studio: directionsFull size
Three generated directions for one brief.
03Canvas

Open it and click around.

The prototype runs in the canvas. It is real HTML, so hover states and animations work.

Possibility Studio: canvasFull size
A prototype running in the workbench canvas.
04Revisions

Say what to change.

Type an instruction like 'make the clock quieter', or point at a part. Each change becomes a new revision.

Possibility Studio: revisionsFull size
Revision history for one prototype.
05Report

Check the report.

See what it says it built, which requirements passed, and what it assumed, before you trust the preview.

Possibility Studio: reportFull size
The requirement report for a saved revision.
06Export

Export and ship.

Download one standalone HTML file with assets inlined. No build step.

Possibility Studio: exportFull size
The source view for a saved revision.
3
Paths per brief
1
Request at a time
20 MiB
Per asset
MIT
License

Install it locally.

Open the repo pip install -e '.[dev]' uvicorn studio.app:app --port 8310
Stated plainlyA provider must be configured before anything can be generated. Browser checks are limited runtime observations, not proof of visual fidelity.