How to Extract Text from Any Image (Screenshots, Scans, Photos)

Screenshots, scanned documents, photos of signs, receipts, book pages, slides captured during a talk — they all have text trapped inside an image. Pulling that text out by hand means retyping, which is slow and error-prone. The faster path is OCR (optical character recognition): a tool that looks at the image and returns the text it sees. This guide is the practical version: which use cases actually work, the accuracy tips that matter, and how to do it in your browser without uploading anything to a server.

When OCR is the right answer

OCR is excellent for:

• Screenshots of text. Error messages, code snippets, paragraphs from articles, the body of an email. These almost always come through cleanly because the underlying text is sharp and the font is well-known.
• Scanned documents. Invoices, contracts, forms, typed letters. The scanner has already cleaned the image up, so the recognition engine sees a clean black-on-white page.
• Photos of printed signs, menus, and labels. The text is meant to be read; the recognition works as long as the photo is reasonably in focus and well-lit.
• Scanned PDFs. A scanned PDF is just a PDF made of images. Render each page and run the same recognition on each one.

OCR is not the right answer for:

• Handwriting. Neat block letters may work; cursive, doctor notes, and writing in unusual languages usually do not. For serious handwriting recognition, a dedicated model trained on handwriting works better, but no in-browser option is reliable today.
• Heavily stylized fonts. Decorative, handwritten-style, or narrow fonts degrade accuracy. Plain sans-serif and serif fonts work best.
• Text on noisy backgrounds. Text over photos, complex gradients, or busy patterns is much harder. Pre-process the image (crop, increase contrast, convert to grayscale) if the result matters.
• Very small or very low-resolution images. Below ~150 DPI equivalent, accuracy drops fast. Re-scan at a higher resolution.

How to extract text from a screenshot

The fastest path: open the Image to Text (OCR) tool, drop in the image, pick the language, click Extract. The whole process runs in your browser — the image never leaves the page. The result shows the recognized text on the right, with a confidence score on each block so you can spot the parts that might be wrong.

Three things to know for best results:

1. Match the language dropdown to the language in the image. Picking the wrong language produces nonsense. If the image is in English, leave the dropdown on English.
2. Higher resolution is better. If you control the screenshot, take it at the native resolution. If you are working from a small image, use the Image Resizer to upscale it 2x before extracting — the recognition model is more accurate on larger text.
3. Crop out the noise. If the image has a lot of background or a sidebar, crop to just the text area first. The Image Cropper works in your browser.

The first time you run the tool, the language model downloads once (~10 MB) and is cached by your browser. Subsequent extracts on the same language are 1-3 seconds for a typical screenshot.

How to extract text from a scanned PDF

A scanned PDF is a PDF whose pages are images, not real text. You can copy text from a real PDF; from a scanned one, you get nothing because there is no text — only pixels. The fix is to render each page to an image and run OCR on it.

The PDF OCR tool does this in one step: drop in the PDF, pick the language, click Extract text from all pages. The tool renders each page to a high-resolution canvas (300 DPI), runs recognition on each one, and gives you the combined text with per-page boundaries.

This is the right tool when:

• The PDF is a scan of a printed document (invoices, contracts, books, government forms).
• You need searchable text from a document you cannot re-export.
• You want to copy text from a PDF someone sent you.

This is the wrong tool when the PDF already has real text. In that case, just open it in a viewer and copy — recognition is unnecessary and may even degrade the result. If you have a mix, the PDF to JPG tool lets you pre-render just the pages you need.

How to extract text from a phone photo

Phone photos are the hardest case because of the lighting, angle, and focus variability. To get clean results:

1. Hold the phone parallel to the text. Perspective distortion is the #1 reason phone-photo OCR fails. A flat, straight-on shot works dramatically better than a casual angle.
2. Get the lighting right. Even, indirect light (near a window, not in direct sun) is the best. Direct flash creates glare and reflections. A shadow across the text is worse than low light.
3. Tap to focus on the text. Most phone cameras autofocus on the center of the frame. Tap the text on the screen to force focus there.
4. Get close enough to fill the frame. If the text is small, get closer rather than zooming — the optical resolution is better than the digital zoom.
5. Steady your hands. A blurry photo is the second most common failure. Lean against something or use both hands.

If the result is still messy, the post-processing moves are: convert to grayscale (improves contrast), increase sharpness, and crop to just the text area. The Image Compressor tool does not help here, but most image editors can do these in seconds.

How to extract text from a printed page (book, manual, magazine)

For a single page, the OCR tool works. Take a photo or scan, drop it in, get the text. For many pages, the PDF OCR tool is the better entry point because you can combine all the pages into one document at the scanner.

Tips for books specifically:

• Scan at 300 DPI or higher. Body text below 9pt is unreadable at 150 DPI.
• Use the right page binding. Lay the book flat and press the spine, or use a V-shaped book scanner. The fold at the spine distorts the text, and the recognition will fail on the inside column.
• Skip the page numbers and headers after the fact. They add noise to the output.

Privacy: why the in-browser approach matters

Most "free" OCR tools upload your image to a server to do the recognition. The server sees every photo of a receipt, every screenshot of a contract, every personal document you scan. The same companies often retain those images to train future models. For a one-off recipe, this is fine. For an invoice with your address, a tax document, or a scan of a passport, it is not.

An in-browser tool runs the recognition engine locally. The image never leaves the page. The only network requests are the initial download of the language data and the static assets of the page itself. No tracking, no model training on your data, no third-party access. The same is true for the PDF OCR tool — the PDF never leaves your device, and the engine is compiled to run in the page.

For sensitive documents, this is the only acceptable approach.

What about handwriting and complex layouts?

Honest answer: today's in-browser OCR is excellent on printed text and poor on handwriting and complex multi-column layouts. For handwriting, you need a model trained on handwriting — those exist, but they are larger, slower, and not yet standard in browsers. For complex layouts (multi-column scientific papers, mixed text-and-figure pages, receipts with multiple sections), recognition often gets the text right but loses the structure: column 2 may end up above column 1, captions get mixed with body text, table cells lose their alignment.

The fix for structure is post-processing: if the output is wrong, paste it into a text editor and fix the order. For most use cases, the time to fix is far less than the time to retype.

Quick checklist for best OCR accuracy

1. Match the language dropdown to the image. Wrong language = nonsense.
2. Use a high-resolution source. 300 DPI for scans, native resolution for screenshots, full-resolution photo for phone shots.
3. Crop to the text area. Less background = less noise.
4. Straighten the image. A 1-degree rotation can drop accuracy dramatically.
5. Convert to grayscale for low-contrast text. Stamps, faded ink, watermarks — grayscale often helps.
6. Check the confidence scores. Low-confidence blocks (under 60%) are the ones to verify by eye.

For most images, the default settings on the OCR tool will get you 90%+ accuracy on printed text. For the rest, the steps above cover the common failure modes.