Screenshot to Text for ChatGPT

ChatGPT's context window

GPT-4o handles 128K tokens. A typical screenshot of a paragraph is 100–500 tokens once OCR'd.

GPT-4 Vision charges ~85 tokens per low-res tile + ~170 per high-res tile. A full-page screenshot in detail mode can hit 1,500+ tokens before you ask anything.

Want exact numbers? Count tokens for ChatGPT →

Image rules to know

If you do upload the image, ChatGPT resizes it so the longer edge is ≤2048 px. Tiny text below 12 px tends to mis-OCR — crop tighter or upscale before sending.

The workflow

1. Drop the screenshot into the OCR tool — Tesseract runs locally and returns clean text in ~2 seconds.
2. Copy the recognized text and paste it into ChatGPT.
3. Ask your question — “Summarize this”, “Translate this”, “Find the bug”.
4. If the screenshot is a chart or UI, send the original image instead — that's where vision actually adds value.

Common pitfalls

• Uploading every screenshot to GPT-4 Vision when text-only OCR would do the job for free.
• Not cropping — extra UI chrome OCRs into junk lines that confuse the model.
• Trusting the OCR for code without a spot-check; punctuation and indentation are common errors.

Frequently asked