Extract text from a screenshot

Screenshot in, text out

You screenshotted something — a Slack error, a slide, a chat, a receipt, a subtitle frame — and now you need the words out of it, not a picture of them. chunks.md reads the screenshot and hands back editable text.

Drop the image onto the page or click to browse, choose a model, and press Chunk it. A few seconds later you have clean, copyable text — as plain text or as markdown that keeps the structure. No retyping, no signup, no waiting on a server.

Your screenshot never leaves your device

Screenshots are rarely just words. They carry private things — direct messages, email threads, dashboards, account numbers, error logs with tokens in them. Most "free online OCR" tools upload that image to a server you know nothing about.

chunks.md does not. The optical character recognition runs on your own device, in the browser, using WebAssembly and WebGPU. Nothing is uploaded, there is no account, and after the first visit it works offline. Sensitive screenshot, sensitive document, doesn't matter — it stays with you.

If a tool asks you to upload a screenshot of anything private, that image is now on someone else's computer. On-device OCR is the only version of this where that never happens.

Sharper than a basic OCR box

Most browser OCR tools wrap a single lightweight engine that struggles with dense or low-contrast screenshots. chunks.md gives you 8 OCR models in one tab, so you can match the model to the shot:

• Fast, everyday screenshots — PaddleOCR v5 reads clean UI text and captions accurately in a couple of seconds.
• Dense, layout-heavy captures — tables, multi-column dashboards, small print: Unlimited-OCR (DeepSeek-based) is layout-aware and runs on WebGPU for complex pages.
• Japanese and manga screenshots — a dedicated Japanese model handles vertical text, furigana, and comic panels that general OCR mangles.

Switch models and re-run in the same tab until the extraction is clean — no re-uploading, because there was never an upload.

More than a copy-paste

Because the output is markdown, not just a flat wall of text, it keeps headings, lists, and tables intact. That makes a screenshot useful somewhere other than your clipboard: paste it straight into notes, a document, or an AI assistant as clean, structured context that ChatGPT, Claude, Gemini, or Perplexity can actually read.

How to extract text from a screenshot

1. Open chunks.md and drop your screenshot onto the page, or click to browse for it.
2. Pick a model — the default is a good all-rounder; switch to a heavier one for dense or non-Latin text.
3. Press Chunk it and let it run on your device.
4. Copy the markdown, or download it as a .md or .txt file.

Everything happens locally. Your screenshot is never sent anywhere.

Common questions

How do I extract text from a screenshot? Open chunks.md, drop your screenshot onto the page (or click to browse), pick a model, and press Chunk it. The text comes back as clean markdown you can copy or download — usually in a few seconds.

Can I extract text from a screenshot without uploading it? Yes. chunks.md runs OCR entirely in your browser using WebAssembly and WebGPU. The screenshot is never uploaded to a server — it stays on your device the whole time.

Is it free? Yes — free, no account, no page limits. The OCR models download once and are cached locally, so repeat use is instant and works offline.

What image formats can I use? PNG, JPG, and WebP — which covers every common screenshot format on Windows, macOS, Linux, iOS, and Android. You can also drop in a multi-page PDF.

Can it read screenshots in other languages, including Japanese? Yes. There are language packs for English, Chinese, Japanese, Korean, Arabic, Hindi, Cyrillic, and more, plus a dedicated Japanese model for manga and comic screenshots.

Related reading