AI image annotation tool

Most annotation tools make you draw every region by hand. Regionly's purple Auto button flips that: AI suggests labeled regions on your image, you refine them with normal drawing tools, and you export exact coordinates — free, in your browser.

How does AI annotation work in Regionly?

1. Drop your image into the editor (or paste it with Ctrl+V).
2. Click the purple Auto button. Optionally type a hint first — "focus on the furniture", "label the buttons" — to steer it. The AI reads the image and suggests labeled regions adapted to what it's looking at: a game room comes back with walkable floor, obstacles, and doors; a floor plan with rooms and fixtures; a UI screenshot with nav, buttons, and content areas.
3. Refine. The suggestions land as ordinary editable regions, so the seven tools — select, freeform, rectangle, circle, polygon, arrow, and point — let you drag vertices, rename labels, delete misses, and draw whatever the AI didn't catch.
4. Hit Done and export.

The AI does the tedious first pass; you keep final say over every region. That order matters — a suggestion you can correct in two seconds beats a blank canvas and beats blind trust in machine output.

How is this different from CVAT or Labelbox?

Platforms like CVAT and Labelbox are dataset factories: they exist to label thousands of images with teams, review queues, ontologies, and training-pipeline exports. If you're producing an ML dataset at scale, that heavyweight machinery is the point.

Regionly sits at the other end. No install, no account, no project setup — open an image, annotate it, and get exact coordinates into an AI prompt, an agent workflow, or a game. It's the lightweight in-browser option for the jobs dataset platforms are overkill for: grounding a ChatGPT or Claude conversation, feeding regions to an automation script, marking collision areas on room art. If plain boxes are your whole job, the bounding-box workflow covers that in three steps.

What do I actually export?

Two formats, both generated from the same regions:

• Versioned JSON — every region in exact pixels and normalized 0–1 coordinates, same-label regions grouped into one entry, notes attached. Built for code and agents.
• Prompt-ready text — plain language plus numbers, made to paste next to the image in a ChatGPT or Claude chat.

You can also save the entire session as a project file and reload it later to keep editing. The formats are documented field by field in the docs.

Does my image get uploaded?

In normal use, no. Regionly runs 100% client-side — drawing, labeling, and exporting all happen in your browser, and the image never leaves it. AI annotate is the one opt-in exception: nothing is sent until you click Auto, and when you do, the image goes out solely for that analysis and is never stored.

Is Regionly free?

The editor is free with no signup: all seven tools, JSON and prompt-text export, project save and load. AI annotate is part of Premium — one $5/month plan that also adds multi-image projects and batch export.