Summarize a screenshot or a photo of text
Zengist reads the text in a screenshot, a photo of a document or a scanned image, on your device, then summarizes it. Drop up to 20 images in one run, mixed with PDFs and documents if you like. Free, no account, and on desktop Chrome, Edge and WebGPU browsers nothing is uploaded at any point.
What happens to an image
An image has no text layer, so the first step is recognition: the picture is read in your browser and the words are pulled out of it. That works well on screenshots, clean photos and straight scans, and less well on blurry, tilted or low-light photos. A photo taken square-on in good light reads almost as cleanly as a screenshot.
The recognized text is what gets summarized. The image itself is never sent anywhere, on any engine. Even on the cloud fallback, only the recognized text is relayed, and it is dropped after the run.
PNG, JPEG, WebP, GIF and BMP are read. A screenshot of a long chat, a photo of a letter, a picture of a whiteboard, a scanned receipt: each is a few seconds of recognition and then a normal brief.
How to do it
- Drop the images on the card, or paste a screenshot straight from the clipboard. Several images that belong together can go in one run.
- Write what you want from them. "What do I owe and by when" on a photo of a bill; "summarize this thread" on a stack of chat screenshots.
- Press the arrow. The brief lists every date, amount and action found, with one line per image at the end.
Good to know
- Up to 20 images per run, 50 MB each. Recognition takes a few seconds per image.
- Handwriting reads poorly; printed and on-screen text read well.
- The image is never uploaded. On the cloud fallback only the recognized text travels.
- Crop to the text before dropping if a photo has a lot of background; recognition is faster and cleaner.
Tips
- Screenshots of a long conversation: number them in order before dropping so the brief follows the thread.
- Photos of a form or a statement: ask for the fields and their values as a list.
- A whiteboard after a meeting: ask for decisions and next steps, not a description of the board.
Questions
Is the picture uploaded?
No. Text recognition runs in your browser. On an on-device run nothing leaves your device; on the cloud fallback only the recognized text is relayed and then dropped.
Which image formats work?
PNG, JPEG, WebP, GIF and BMP, plus anything you paste from the clipboard as a screenshot.
How accurate is the text recognition?
Very good on screenshots and clean scans, good on square photos in decent light, poor on handwriting and blurry shots. The brief is only as good as the recognized text, so check anything that matters against the picture.
Try it now. Also: pdfs, word documents, web pages. Every question in one place in the FAQ; stuck on something, see help.