Scanned document translation
In a scan, the text is not text — it is a picture. You cannot select it, and ordinary translation tools cannot read it, which is why dropping a photocopied archive into a translation box usually returns nothing at all.
Geyi recognises these files before translating them: it reads the characters out of the image, works out the reading order, then translates and lays the result out again. Multi-column pages come back column by column rather than scrambled, and vertical text is converted to horizontal.
Which originals belong here
- Scanned book pages and photocopied archives — PDFs produced by a scanner or copier.
- Photographs of documents — PNG, JPEG, WEBP, TIFF and similar image formats.
- Image-only PDFs — the kind where no text can be selected in a reader.
- Older publications with complex layouts — multiple columns, vertical setting, two-page spreads.
When recognition is most accurate
Clean print scanned straight gives the highest accuracy. Blurry, skewed or unevenly lit images, and pages with a lot of handwriting, are markedly harder — for those, check the key passages against the original before exporting.
Vertical Japanese, two-page spreads and other unusual reading orders are corrected automatically, with no need to split pages or reorder anything yourself.
When you only want the original text
If all you need is the text out of the scan, without translation, document extraction fits better: it bills 4 credits a page and produces horizontal, editable Word, PDF, TXT or Markdown in whatever language the original was.
Extracting first to check recognition quality, then deciding whether to translate, is also a way to spend fewer credits.
Frequently asked questions
What does translating a scan cost?
Scans go through the recognition path used by AI reflow: 8 credits a page on high precision, 4 on standard, with a 10-credit minimum per file. A single image counts as one page.
Why is it more expensive than a normal PDF?
A scan has to be recognised before it can be translated, and that step costs considerably more compute than reading an existing text layer. Digital PDFs on layout-preserving mode are 4 credits a page.
Do multi-column layouts come back scrambled?
No. Multi-column pages are restored in reading order rather than swept left to right into one block.
Can handwriting be recognised?
Printed text is recognised far more reliably than handwriting. Where a document is largely handwritten, results need checking — run document extraction first to see the recognition quality.
What format is the finished translation?
Scans go through AI reflow, so you can export Word, PDF, Markdown or EPUB. Note that the translation is not painted back onto the original image; it is laid out as an editable document.
Does a single image really cost 10 credits?
Yes — every file bills at least 10 credits and one image counts as one page. Combining several images into one PDF before uploading bills on the total page count, which works out cheaper.
Related
中文版
Geyi is a proprietary AI document platform owned and operated by Nayuta Technology and Culture Company Limited.