Press a picture. Open the page behind it.
Name a subject. You get one annotated plate — six to eight labelled zones, callouts carrying real dates and dimensions, a caption written like a museum placard. Press any zone and a new page opens about whatever is drawn there. Keep pressing and you have a tree of encyclopedia plates you navigate by pointing at pictures.
Both examples ship with saved model output for every lane — the plan, the transcribed text layer, the drill and the placard — so you can walk a finished canvas without signing in and without spending a credit. Opening your own canvas is free too: a planned page is a real diagram, and pressing a planned zone opens its child page for nothing. Drawing a plate, transcribing one, asking the model about a press, and writing a placard are metered runs, and each one is a button you press yourself.
How it works
- Plan — the subject becomes a plate: zones with boxes, headings, callouts with numbers, a caption, a compiled prompt, and the facts it drew, each marked for confidence.
- Explore for free — the planned plate renders as an SVG diagram with selectable text. Press a zone and its child page opens with the label and brief the plan already wrote. No run, no credit.
- Draw — when a page is worth illustrating, the image model draws the plate from the compiled prompt and the canvas's visual system. It is stored to your account files.
- Annotate — a vision pass reads the labels the drawing actually contains, places them, and says zone by zone what was dropped, what was invented and what is garbled.
- Drill — a long press draws a red crosshair on the plate and asks the model what is under it. The answer names the subject, places its card, and briefs the child page.
- Read — the placard: sections per zone, a glossary, the numbers with their confidence, and the questions this page leaves open — each of which opens a child page.