← All toolsLive demo

Jev Pixels

Type a prompt and Jev paints it pixel by pixel — every pixel one typed color decision, all in parallel. Watch the picture appear, then download it. No key, just paint.

Each of the 120 pixels is one Jev decision — a color from the palette — all run in parallel, streaming in as the picture appears.
0/120pixels
0.0selapsed
0/sdecisions/sec
—Jev cost

Image generation, reframed as a grid of typed decisions. Get an API key →

A playful demo — 120 independent Jev choice calls over a 12-color palette, in parallel. Low-fi on purpose; it shows Jev deciding at scale, not a diffusion model.

Get an API key →API quickstart
Build this with Jev

The same demo is one Jev call

  1. Split a picture into a grid of pixels
  2. For each pixel, ask Jev a palette choice
  3. Run all the pixels in parallel
  4. Paint each cell the color Jev picked
one call per pixel · all parallel · a full grid in seconds
const results = await Promise.all(items.map((item) =>  // items = pixels {r,c}
  fetch("https://jevtypesafeai.com/api/v1/decide", {
    method: "POST",
    headers: { Authorization: `Bearer ${process.env.JEV_API_KEY}` },
    body: JSON.stringify({
      state: item,
      questions: {
        color: { type: "choice", instructions: "Color of this pixel in a picture of '${prompt}'?",
                 criteria: palette },   // { red:"…", blue:"…", … }
      },
    }),
  }).then((r) => r.json())
));
paint(px, results[i].answers.color.choice);   // → "red"
See a real decision RED · 181ms
State
A 12x10 pixel-art picture of: "a red apple on white".
What color is the pixel at row 4, column 6?
Question choice · color
red86%
green8%
white6%
latency 181mscost $0.000009

How it works

This isn't a diffusion model — it's a decision model painting. The picture is a grid, and each pixel is one independent Jev call: given the prompt and the pixel's position, Jev picks a color from a small fixed palette. All the pixels run in parallel and stream back, so you watch the image resolve out of a hundred-plus tiny typed choices in a couple of seconds.

It's deliberately low-fi. The point isn't photorealism — it's to show a fast, cheap decision model making a coherent choice at every position, the same primitive that routes requests and gates tools, here arranged into a picture.

Why a decision model can do this

Each call returns a typed choice with calibrated probabilities — no free-form output to parse, no chance of an invalid color. Because every pixel is bounded and independent, they parallelize perfectly: a 120-pixel image is 120 calls that finish together. Swap the palette or the grid size and the same pattern scales.

FAQ

Is Jev an image generator?

No — Jev is a decision model. Here we reframe image generation as a grid of per-pixel color choices so you can see it decide at every position. It's a toy that shows the primitive, not a replacement for a diffusion model.

Why is it so low-resolution?

Each pixel is a separate API call, so the grid is kept small (120 pixels) to stay fast and free. The charm is watching a recognizable picture emerge from that many tiny typed decisions.

Is it free? Can I download it?

Free, no key — a rate-limited batch of calls to the live Jev API. When it's done you can download the result as a PNG.

More: Jev Wikiracer · Ask 100 Personas · All Jev tools · Jev API docs

Jev Pixels — Jev paints a prompt one pixel at a time · Jev by TypeSafe AI