How it works

Four steps, and only one of them is the AI.

The generation is a starting point, not the deliverable. Everything that makes a screen good still happens where it always did — in the canvas, with your hands on it.

Describe the screen

Write the brief the way you would say it to a colleague. Name the surface, the sections and the mood — the model reads all three.

Pick the device first; width is treated as fixed and height grows to fit.
Naming sections ("hero, three features, footer") reliably produces them.
Mood words steer palette and type more than they steer layout.

Watch it draw

The scene arrives as real objects, placed back to front so backgrounds land before the content that sits on them.

Every element is validated and clamped to the frame before it is drawn.
The whole generation is a single undo step — Ctrl+Z removes all of it.
Between 24 and 80 elements for a landing page, fewer for a single view.

Take the pen back

Nothing is flattened. Select, resize, recolour, regroup, mask and rewrite exactly as you would in a file you drew yourself.

Corner handles scale type with the box; edge handles crop it.
Drag into a frame and the layer reparents itself automatically.
Smart guides measure the gap to the nearest neighbour as you move.

Ship the asset

Export a single frame or the whole canvas as PNG or PDF, at full resolution, without moving the work somewhere else to finish it.

Frame export renders only that frame's descendants, on its own background.
Fonts are awaited before rasterising so type never falls back mid-export.
PDF is written directly, with the image embedded rather than re-encoded.

Editorial — a screen composed in Atelier

It takes one sentence to find out.

The free plan is enough to see whether this fits how you work.

Start creating