How to Extract Text From a Photo (No Upload)

Screenshot OCR is the easy case: flat pixels, straight lines, no lighting to worry about. A photo you take with your phone is not that. Point your camera at a receipt, a business card, a whiteboard, or a page in a book and you get a skewed angle because you shot it from wherever you were standing, glare off glossy paper or a laminate card, shadows from your own hand or the phone itself, and curled or creased paper that bends text out of a straight line. None of that is a bug in the photo — it's just what a real-world camera shot looks like, and it's the reason photo OCR needs to be more forgiving than screenshot OCR. Our Image to Text tool is built for exactly this: real photos of real objects, not clean screen captures.

Open the Image to Text tool, drop in the photo straight from your camera roll, and let it run — there's no account, no upload progress bar to wait on, and no size limit to worry about for a single photo. The tool recognizes the text in the browser and hands you back editable, copyable plain text you can paste into a note, spreadsheet, or email. If the photo has multiple blocks of text (a whiteboard with several sections, a page with a caption), you'll get all of it back in one pass rather than needing to crop and re-run for each block. Two rounds are worth trying if the first pass looks rough: retake the photo straighter and better lit, or straighten and crop the existing shot before re-running it, which is often enough to turn a partial read into a clean one.

OCR accuracy tracks photo quality closely, and the three fixable culprits are angle, light, and paper shape. For angle, shoot as close to straight-on as you can, or use Rotate and Crop afterward to square up the frame and cut out background clutter the engine doesn't need to see. For glare on receipts, laminated cards, or glossy pages, tilt the paper or your phone slightly rather than shooting dead-on under an overhead light — direct reflections wash out the exact pixels the text sits on. For shadows, take the photo with even light across the whole page rather than a single side-lit lamp, and flatten curled receipt paper or a book's page curve as much as you can before shooting. None of this needs to be perfect — the engine tolerates real-world imperfection — but a straighter, better-lit shot is the single biggest lever you control.

If you're on an iPhone, your camera roll is probably saving photos as HEIC by default, not JPG. Image to Text decodes HEIC/HEIF automatically using the same decoder every other single-image tool on pixeltools already relies on, so you can drop the file in exactly as it came off your phone — no manual "convert to JPG first" step, no separate converter tab. If you'd rather have a JPG copy of the photo for other reasons, Image to Text doesn't require it, but pixeltools' HEIC to JPG converter is there if you want one.

Photo OCR earns its keep on the stuff you can't screenshot: a paper receipt you want logged before it fades, a business card someone handed you at an event, notes scrawled across a whiteboard after a meeting, a page from a physical book or printed document you don't want to retype, a street sign or menu in a language you want to search or translate, and handwritten notes on paper (results vary more here — tidy print reads far more reliably than cursive). In every case the text existed on paper or a surface first, a phone camera was the only practical way to capture it, and the goal is the same: get it into plain, searchable, copyable text without retyping it by hand.

Tesseract, the recognition engine behind the tool, doesn't just return nothing when you feed it a photo of wood grain, fabric, or a blank wall — it hallucinates plausible-looking garbage characters instead, which is worse than an empty result because it looks like a real (wrong) answer. Image to Text catches this using the engine's own confidence score. In our testing, pure-noise hallucinations scored around 13 out of 100, genuinely blurry-but-real text still scored 65 or higher, and a clean, well-lit photo typically scores 90+. The tool uses a threshold of 35 to flag the no-real-text case without rejecting photos that are just honestly hard to read, so a low-confidence result is a meaningful signal to retake the photo, not a false alarm on a tricky one.

Image to Text runs on the same self-hosted tesseract.js engine and language files as pixeltools' OCR PDF tool — same LSTM-only recognition mode, no third-party CDN, English only for now. If you're deciding between the two: use Image to Text for a photo file (JPG, PNG, HEIC, WEBP) and OCR PDF when what you actually have is a scanned or photographed PDF you want to keep as a searchable PDF, not just plain text. Either way, recognition happens entirely client-side — your photo is never uploaded to a server, which matters if it's a receipt with card digits on it or a business card with someone's personal contact details.