A structured measurement of a reference image: text with positions, a colour grid and a composition reading.
About
Reads a reference image (PNG, JPEG or WebP) and returns structured measurements: the text it finds with each line's position, an 8 by 8 grid of sampled colours matched to the brand's own colour tokens where they are close, and one vision reading of the composition, the type hierarchy, any gradients and every logo or brand mark, so a recreation can leave them out.
It is for rebuilding a design you like in a different brand rather than copying it. The text and colour measurements are taken locally and handed to the vision reading as ground truth.
The reading is a paid call to a vision model. On very dense images, text recognition stops after about 25 seconds and the result says so; the colour grid and the reading still come back.
- TypeScript
- Tesseract OCR
- AI vision models
Requirements
- Runs with
- Node.js 20 or later
- Operating systems
- windows, macos, linux
- Network
- Sends the image to the vision provider over the internet. The first run downloads English text-recognition data.
Costs
- External costs
- OpenAI: billed to your own key — Planned direct lane: you would bring your own OpenAI key. Today the call runs through the Syvon provider proxy and Syvon credits.