OCR for AI inputs.

Drop a screenshot. Get clean, AI-ready text. Browser-side OCR — your image stays on your device.

When you only have a screenshot — of an error message, a chart, a slide, a scanned page — you can't paste pixels into a text-only chat. Screenshot to Text runs OCR right in your browser via Tesseract and gives you selectable, copyable text. No upload, no account, no API key.

How to use it

  1. 1.

    Drop the image

    Supports PNG, JPG, WebP, GIF, BMP, and TIFF. Multi-line screenshots and full-page captures both work.

  2. 2.

    Wait for OCR

    First run downloads a ~10 MB language model (cached afterwards). Subsequent images are much faster.

  3. 3.

    Use the extracted text

    Copy the result, paste it into ChatGPT or Claude, and ask follow-up questions about content the model couldn't see directly.

When OCR beats vision models

Multimodal models can read images, but for dense text — code, contracts, tables — OCR is often more accurate, much faster, and uses no API quota. It's also the only way to feed image-only content into text-only models or older API tiers.

Best for

  • Error messages from screenshots
  • Slides and presentation captures
  • Text in screenshots of articles or tweets
  • Quickly extracting code from a screenshot

Pixels to prose.

Tesseract.js runs entirely in your browser, recognizing English text from PNG, JPG, WebP, and other common formats. We then strip very short UI fragments and rebuild paragraphs so the output reads cleanly when you paste it into ChatGPT, Claude, or NotebookLM.

Frequently asked

Related reading