How it uses Jev
A prompt and selected representation/grid settings are converted in the browser into batched Jev questions about palette, HSL, RGB, or silhouette values. The browser combines returned probabilities and samples them into a rendered painting; a local Python helper forwards API requests because the endpoint does not allow browser CORS.
What Jev decides
Is pixel (x=0, y=0) inside the foreground silhouette described in state?
Image description, grid size, pixel (0,0) and silhouette task.
App workflow
- Set up a canvas request
The user enters a prompt, chooses a representation and grid size, and starts a painting in the browser.
- Ask in batches
Browser code constructs and batches the representation-specific questions and sends them through the loopback helper to Jev.
- Transform probabilities
The browser combines channel probabilities where needed and uses spatially correlated sampling, mean color, and entropy to derive pigments and relief.
- Render and retain
A module worker computes the image and adds it to the in-memory carousel; reloading clears paintings.
Why it is interesting
It treats Jev probability distributions as generative material: spatial sampling, entropy, contours, and lighting turn typed pixel-level outputs into textured artwork rather than a conventional answer display.