Screenshot OCR · for Claude
Screenshot to Text for Claude
Convert screenshots into text Claude can reason over — useful when you want exact strings, codeblocks, or to skip vision-token overhead.
Open Screenshot to TextClaude's context window
Claude 3.5 Sonnet supports 200K tokens of input. A typical screenshot's worth of OCR'd text is well under 1K.
Claude bills images at roughly 1.15 tokens per pixel area in tiles. A full-screen 2560×1440 screenshot can cost 1,600+ tokens just to attach.
Want exact numbers? Count tokens for Claude →
Image rules to know
If you do attach the image to Claude, the recommended max edge is 1568 px (anything larger is downscaled). Aspect ratios beyond 2:1 hurt accuracy.
The workflow
- Drop the screenshot into the OCR tool — runs locally, no upload.
- Paste the text into Claude with a short instruction (“Translate this”, “Refactor this snippet”).
- For mixed text + visual content, send both: paste the OCR text and attach the image, then ask Claude to use both.
- Spot-check OCR output before trusting it for code or numeric data.
Common pitfalls
- Cropping out the title bar but leaving the taskbar — extra OCR noise.
- Sending only the image when you could have sent text — costs more and Claude's answer will be the same.
- Trusting OCR for monospace code; double-check brackets and quotes.
Tool
Screenshot to Text
OCR screenshots into AI-ready text.
Frequently asked
Is Claude's vision better than ChatGPT's?
They're close. Claude tends to be more cautious about ambiguous visuals; ChatGPT is more willing to guess. For plain text, OCR sidesteps both.
Can I attach OCR output as a Project file?
Yes. Save the cleaned text as a .txt or .md file and attach it to a Claude Project for re-use across sessions.
Does Tesseract handle non-English text?
Yes — language packs cover 100+ languages. The tool defaults to English; switch languages in the UI.
Screenshot OCR for other models