How to Extract Text From a Photo With OCR, Cleanly
A photograph of a notice board, a handwritten-free printed form, a textbook page: OCR can turn any of these into copyable text in seconds. It can also turn them into confident nonsense when the photo is dark, tilted or cropped through a line. The difference is almost never the engine; it is the ninety seconds of prep before you press the button.
What OCR actually reads
Optical character recognition looks for contrast edges shaped like glyphs. It loves flat, evenly lit, high-resolution prints with dark ink on light paper. It struggles with gloss reflections, shadows across a line, decorative typefaces, low contrast (grey on cream), and tilt beyond a few degrees, because every one of those deforms the edge shapes it matches against.
Handwriting is a different sport: general OCR handles neat block capitals partially and cursive barely. If the source is handwritten, expect keywords rather than sentences, or re-write the parts that matter.
Ninety seconds of prep that doubles accuracy
- Light: daylight from behind you, no flash; flash creates a hotspot that erases ink.
- Angle: square to the page, phone parallel; tilt stretches glyphs into cousins of themselves.
- Distance: fill the frame with the text block; a page occupying a third of the photo loses small type.
- Crop: trim to the text region before OCR, margins and tables confuse line order.
- Flat: press books and receipts flat; a curved page reads as two slopes.
If the photo already exists and is bad, re-shoot rather than fight: prep is cheaper than proofreading.
Running OCR in the browser
An in-browser image-to-text tool keeps the photo on your device, which matters for medical slips, salary documents and ID-adjacent papers. Drop the image, choose the language if offered, and let the on-device model run; first runs download the model once, later runs start instantly.
Copy the result into a plain text buffer first, not straight into your destination. The buffer is where the next step happens, and skipping it is how OCR errors reach published documents.
OCR accuracy is decided by the camera, not the engine.
The proofread pass: where OCR lies
Errors cluster in predictable shapes: zero for letter O and the reverse, one for lowercase l and capital I, rn read as m, apostrophes eaten, and in Devanagari, matras shifted or dropped so a word changes meaning silently. Numbers deserve a second look always: a fee of 1,500 read as 1.500 is a different universe.
Read the extracted text against the photo once, line by line, with a finger or cursor. For anything legal or financial, read the numbers aloud; your ear catches what the eye slides over.
When to switch to PDF OCR instead
One page lives in a photo; ten pages live in a PDF. Scanned documents, book chapters and multi-page forms belong in a PDF OCR workflow that keeps page order and can emit searchable PDF or Markdown. Photos suit notices, receipts, whiteboards and the odd textbook page.
And keep the source photo until the extracted text has been used once. Re-running OCR after fixing the photo is the cheapest upgrade path there is.
Frequently asked questions
Does image OCR upload my photo?
Not with an in-browser tool: the model runs on your device after a one-time download, and the photo never leaves it.
Why is Hindi text coming out wrong?
Devanagari matras are small and OCR drops or shifts them. Shoot sharper, crop tighter, and proofread words rather than glancing at lines.
Can OCR read handwriting?
Block capitals partially, cursive barely. For handwritten sources, treat output as keywords and re-type what matters.
We build and test free, in-browser PDF and image tools. Every guide is written from hands-on testing with real files, and updated whenever browsers or formats change.
Image to Text toolAll articles →Do it now, in your browser
Nothing is uploaded. The file is processed on your device and never leaves it.
Open Image to Text →