Guide
What is the AI OCR tool?
This AI OCR transcribes a photo, screenshot, or scan into structure-preserving Markdown: headings, bullet lists, and tables are reconstructed, and handwriting is supported. Anything the AI cannot read is marked inline as [illegible] instead of guessed. One image costs 3 credits on paid plans (20 credits on the free plan), and the free monthly credit allowance covers occasional use.
Here is how it compares to the free OCR:
| Extract text from an image (free) | This tool (from 3 credits) | |
|---|---|---|
| Engine | PaddleOCR in your browser | AI vision model |
| Printed text | Good | Good |
| Handwriting, whiteboards | Poor | Good |
| Tables, headings, lists | Plain text only | Reconstructed as Markdown |
| Multi-column layouts | Reading order may flatten | Linearized in reading order |
| Where the image goes | Never leaves your device | Private workspace storage |
| History | None | Saved per run |
When it comes in handy
- Turning a whiteboard photo into meeting minutes: the headings and bullet structure come through as Markdown.
- Making a paper document with tables copy-pastable: grids are reconstructed as Markdown tables.
- Importing paper documents into Markdown-based tools like Notion or Obsidian.
- Reading handwritten memos: rushed writing works too (there is also the dedicated Transcribe handwriting with AI).
- Fixing what free OCR flattened: multi-column pages are linearized in proper reading order.
How to use
- Upload an image (on a phone, tap "Take a photo" to shoot it right there). PNG, JPEG, and WebP up to 10MB work; the image is stored privately in your workspace.
- Click "Transcribe". A run takes well under a minute and is saved to your history.
- Review the Markdown preview side by side with the source image. Unreadable spots appear as [illegible].
- Copy the Markdown, or download it as a .md or plain .txt file.
For example, a whiteboard photo from a planning meeting comes back as a ## Next sprint heading, the action items as a bullet list, and the owner/deadline grid as a Markdown table. One smudged word is marked [illegible], so you know exactly what to double-check.
Notes
- The output is a faithful transcription, not a pixel-perfect layout reproduction. Multi-column pages are linearized in reading order.
- Very low-resolution or heavily blurred photos reduce accuracy; the [illegible] markers show where.
- Always check the marked spots against the source image before using the text somewhere important.
- For plain printed text, the free Extract text from an image may be all you need. For business cards, see the business card tools.
