📝 Image to Text (OCR)

Step-by-step

1. Pick an image with text Screenshots, scans, photographed documents, signage — all work. The sharper, straighter, and larger the text, the better the result, and that matters more than any setting here.
2. Choose the right language This is the setting with the biggest effect on accuracy. Choosing the wrong language drops recognition to nearly unusable, because the engine matches against that language's character models.
3. Decide on pre-processing Greyscale suits most cases. Black and white thresholding is markedly better for dark text on a light background (scans, printed pages) but destroys anything photographic. When unsure, start with greyscale.
4. Press start The first run downloads the engine and language data (around 12MB for Chinese), so expect a wait. The browser caches it afterwards and subsequent runs on the same device are much faster. Progress appears in the status line.
5. Copy it or download the .txt The result appears below with an average confidence figure. Below about 70% confidence, proofread carefully. Copy it directly or download it as a plain text file.

When you would use this

Four things that determine accuracy

How well OCR performs is mostly about what you feed it, not how you configure it. In order of impact:

1. The right language — the engine matches against that language's character models. Running Chinese through the English model returns gibberish.
2. Large enough text — at least 20 pixels of character height, 30 to be comfortable. This is the most common cause of failure, and phone screenshots easily fall below it after scaling.
3. A straight image — a few degrees of skew measurably hurts accuracy. Shoot documents square on, or straighten them afterwards.
4. Strong contrast — dark text on a light background is ideal. Coloured backgrounds, gradients, and shadows all interfere.

Greyscale or thresholding

Greyscale simply removes colour while keeping every level of light and shade. Safe and general — use it when unsure.

Black-and-white thresholding forces every pixel to pure black or white, using the image's mean brightness as the cut-off. For printed pages this is a powerful clean-up: background noise, paper texture, and light shadows all disappear, leaving only the text. But if the image contains photography, a gradient background, or text that is itself pale, thresholding destroys the information outright.

How this differs from online OCR services

Most online OCR uploads your image to their servers, recognises it there, and sends the text back. That means your receipts, business cards, and documents have passed through somebody else's machine.

This page uses the WebAssembly build of Tesseract, so the entire recognition happens in your browser. Only the engine and language models travel over the network (downloaded once from a CDN), and they are unrelated to your image content. The trade-offs are a first-run download and accuracy below the large cloud models commercial services use — an explicit exchange that is usually worth it when the documents are sensitive.

Frequently asked questions

What you might need next

Image work rarely ends in one step. These pair up with Image to Text (OCR) most often:

• 📑 PDF to Images — Every page as a JPG or PNG
• ✂️ Crop Image — Drag to select the area to keep
• 🔍 Upscale Image — Enlarge up to 4× the original size

Want to share the result? Upload turns images, photos, or videos into a short link with optional password, expiry, and view limits. Or head back to all 12 tools.