ocr

Florence-2 runs on-device (WebGPU, or WebAssembly as a fallback). The image never leaves your device.

drop an image, or click to choose

stays on your device

task

result

Florence-2 is a compact vision-language model; here it runs entirely in the browser through Transformers.js. OCR pulls selectable text out of screenshots and photos; captioning describes a scene. The first run downloads the model once and caches it.