Table recognition, on your device
Image to Excel
Photograph or screenshot any table and get it back as a spreadsheet. The rows and columns are reconstructed from where the words actually sit on the page, and the whole thing runs inside your browser — no upload, no account, no limit.
Drop a photo, screenshot or scan of a table
JPG, PNG, HEIC-converted images or a PDF page. Read on your device — never uploaded.
Never uploaded
The photo is read by this page on your own processor. Nothing is transmitted or stored.
Borderless tables too
Columns are found from the whitespace between them, not from grid lines that may not exist.
Editable before export
Fix a cell or drop a row with one tap, then save a real XLSX workbook or a UTF-8 CSV.
How it works
- 1
Photograph the table straight on
Fill the frame with the table and keep the page flat. A straight-on shot in even light is worth more than any amount of processing afterwards.
- 2
Drop the image here
JPG, PNG, a phone screenshot or a single PDF page. Recognition runs on your own device — the image is never uploaded.
- 3
Check the grid
The columns are found from the vertical whitespace the table leaves between them, so borderless tables work too. Every cell is editable in place.
- 4
Export to Excel or CSV
The first row becomes the header row of the workbook, ready for filters and formulas.
Common questions
Does this work on a table with no borders?
Yes, and this is where most converters fail. Instead of hunting for grid lines, the columns are found from the whitespace: any vertical strip of the image that no word in any row touches is treated as a column gap. Ruled tables, borderless tables and hand-drawn layouts all work as long as the columns line up visually.
Is my image uploaded to a server?
No. Character recognition runs inside this page using WebAssembly on your own processor. That matters for the things people usually photograph — payslips, invoices, medical results, price lists — none of which belong on a stranger's server.
How accurate is it?
On a sharp, straight-on photo of printed text, very. An average confidence score is shown after each read, and every cell can be corrected before export, so a small misread costs you one tap rather than a re-run.
Can I use a screenshot instead of a photo?
Screenshots are ideal — they are perfectly flat, evenly lit and pixel-sharp, so accuracy is usually near perfect. This is the fastest way to get a table out of a website, a dashboard or a PDF viewer that will not let you select text.
What about a full multi-page PDF?
This page reads the first page of a PDF. For a PDF full of transactions, the bank statement converter is purpose-built and reads every page, and for plain text extraction the OCR studio handles unlimited pages.