Choose one image
Import a screenshot, scan, product image or photo, then rotate or enable gentle scan enhancement only when needed.
ImgIng / Image OCR
PP-OCRv6 · LOCAL OCRRecognize Chinese, Japanese and Latin-script text locally
Your image and recognized text are not uploaded. The current unified models cover Simplified and Traditional Chinese, English and major Latin-script languages; Professional and Ultimate also cover Japanese.
Chinese · Japanese · Latin scriptsThree PP-OCRv6 tiersLayout-aware TXT / MarkdownNo image upload
Capability quick facts
These facts describe the current product, not unshipped roadmap work.
- Processing
- Dedicated browser-local Worker; image and text never upload
- Languages
- Fast: Chinese and Latin scripts · Professional / Ultimate: plus Japanese · Korean not included
- Tiers
- PP-OCRv6 tiny / small / medium
- Outputs
- Layout-aware copy, TXT with tabs, Markdown tables and JSON with geometry
Reviewed by the DataDance product and engineering team · Published · Updated
Finish in three steps
Confirm important settings before export; the processing location is always disclosed.
Choose a recognition tier
Fast is the mobile tier; Professional is the desktop default and adds Japanese; Ultimate targets difficult small text.
Proofread and save the layout
Click a polygon to find its line. Copy or TXT keeps visual rows and separates detected table columns with tabs; Markdown emits real tables where structure is stable.
Why three PP-OCRv6 tiers?
Fast minimises download and memory for clear screenshots and mobile devices. It covers Chinese, English, German, Spanish, Portuguese, French and other supported Latin-script languages. Professional balances photographed text and scans and adds Japanese. Ultimate is desktop-only and spends more download, memory and time on difficult small text. Every tier pins the model name, byte size and SHA-256 so the engine cannot silently change underneath a result.
What does ImgIng add around the model?
The source image is preserved, with manual rotation and optional gentle scan enhancement instead of forced binarisation. After recognition, geometry clusters text into visual rows, reconstructs reading order and detects stable table columns. Copy and TXT use tabs between columns, Markdown emits table syntax, and JSON retains corrected text, polygons, confidence and layout rows. The engine runs in its own Worker; changing tiers, clearing or exiting terminates it to reduce retained memory.
Which site languages can the current models recognize?
Fast supports Simplified Chinese, Traditional Chinese, English, German, Spanish, Portuguese and French. Professional and Ultimate add Japanese. The current unified PP-OCRv6 models do not include Korean recognition, so the Korean interface states that boundary instead of silently returning unreliable output.
What remains outside the promise?
Clear print, screenshots, scans, labels and mixed layouts are the primary scope. Handwriting, formula semantics and arbitrary merged-cell tables are not claimed. Low-confidence output stays visible for proofreading instead of being silently rewritten.
Runtime and model files download to your browser, but the source image and recognized text are never uploaded. Mobile exposes only the Fast tier and limits oversized input to reduce the risk of WeChat or Safari closing under memory pressure.
Updated 2026-08-25 · Live capability detection inside the tool is authoritative
Frequently asked questions
These visible answers match the current product behaviour and structured data.
Does OCR upload my image?
No. Decoding, detection, recognition, layout reconstruction and proofreading all run in the current browser.
Which tier should I choose?
Use Fast on mobile, Professional by default on desktop or for Japanese, and Ultimate only for difficult small text or many low-confidence lines.
Does it support every language shown by the website?
Fast covers Chinese, English, German, Spanish, Portuguese and French. Professional and Ultimate add Japanese. The current unified model does not recognize Korean.
How are tables copied?
ImgIng reconstructs visual rows from text geometry. Copy and TXT insert tabs between detected columns, Markdown emits table syntax when consecutive rows share a stable column count, and JSON preserves geometry for further processing.
Explore ImgIng
Each task page documents real settings, limits and format advice—not keyword-swapped duplicates.