Text inside a picture cannot be selected, searched or edited. ToolGPT's OCR reads printed English from photos, screenshots and scanned PDFs in your browser and returns the words as a plain text file.
Read one image, several photos of a document in page order, or every page of a PDF.
Each request reads up to 20 pages or photos. A longer PDF is refused with an explanation rather than cut short.
Keep the line breaks as printed, or join each paragraph's lines for pasting into a document.
The assistant is told how sure the recognition was, and warns you when the text may be unreliable.
The result is a plain text file to download, and the assistant can quote from its first 2,000 characters.
Add a photo, a screenshot or a scanned PDF with the + button. For pages photographed one by one, attach them in order.
Say “image to text”, “OCR this” or “what does this say?”. The job runs in the queue and shows its progress.
When the job finishes, download the .txt file or ask the assistant about what it says.
OCR works best on clean, printed English: letters, receipts, labels, slides and screenshots with text at a readable size. Images larger than 4,096 pixels on the long side are scaled down before reading.
It is not a good fit for handwriting, blurry or tilted photos, decorative fonts, or pages laid out in columns and tables, where the reading order has to be guessed. Recognition covers English only for now, and it does not translate.
ToolGPT is told to wait for the recognized text before it says what a page contains. When confidence is low, it tells you the text may be wrong instead of presenting it as a clean copy. If nothing readable is found, the job fails and says so rather than handing you an empty file.
The OCR engine, Tesseract, runs in your browser, so the image or PDF stays on your device. So that the assistant can answer questions about the result, the first 2,000 characters of the recognized text are shared with the AI in your chat, just like a message you type. Keep that in mind before scanning a sensitive document.
The first OCR job in a session takes a little longer while the engine loads; later ones start straight away.
Only English for now. Text in other languages comes out wrong or empty.
Not reliably. It is built for printed text, and ToolGPT flags low confidence when the result is doubtful.
Ask ToolGPT to save the pages you need as high-resolution images, then read them in groups of up to 20.
No. The image or PDF is read in your browser; only the start of the recognized text is shared with the assistant.
Try it with your own file
Open ToolGPT, attach a file, and type what you want done.