Read text from an image without retyping it
This is an online OCR workflow with Markdown output. The converter uses a vision model to read visible text, then represent its structure: a document heading can become # Meeting notes, a checklist can become bullets, and a code screenshot can become a fenced block. You download a .md text file, rather than an editable copy of the image layout.
Give OCR a legible source
Use an upright, sharply focused image. Crop unnecessary background, keep the entire passage visible and avoid glare. The server normalizes images to a longest side of at most 1,568 pixels, so very small text spread across a large screenshot can become difficult to read. A tighter crop can help, provided it keeps the surrounding context.
For a scanned document already saved as PDF, use PDF to Markdown. This image panel accepts PNG, JPG/JPEG, WebP, BMP, GIF, HEIC/HEIF and TIFF.
Text recognition and data interpretation need different checks
- Printed text and handwriting: compare names, amounts, dates and any unclear words.
- Tables: inspect headers, row alignment, empty cells and decimal separators.
- Code: compare indentation, quotes, punctuation and variable names before running it.
- Equations: review the LaTeX source against every visible sign and symbol.
- Charts: values can be estimated from axes when labels are absent. An estimated value is not an exact transcription.
- Diagrams: the result describes visible labels and connections; it does not recreate the drawing.
The model is instructed not to invent text, but OCR can still omit content or produce plausible errors, including on blank or unclear images. Do not treat a successful conversion as proof of accuracy. Keep the image beside the output during review.
The scan preparation guide explains recognition problems. If you need another language, review the extracted text first, then use Markdown translation as a separate step.