← All models · Scene text detection
Practical · object-detection (text) · WASM
A receipt, a screenshot, a form — detect every text region and black it out before the file ever leaves your machine. The image is never uploaded; redaction is a local canvas operation on the detected boxes.
The redacted image is produced entirely in-page from the detected boxes — nothing is uploaded. This blacks out every detected region; a real tool would let you keep the parts you want.
You don't need to read a receipt to hide it — you only need to know where its text is. Skipping recognition means the sensitive characters are never decoded at all: the model outputs regions, the page fills them with black, done. Cheaper, faster, and it leaks nothing.