Loading
Loading tool
Preparing the tool. The remaining time is not yet known.
All processing for this tool happens in your browser. Your input is not sent to a server.
About this tool
Recognize printed English, Simplified Chinese or Japanese locally with a selected language model. Engine/model assets (about6–8 MiB per run) load from this site; your image is never uploaded. No remote OCR fallback.
Common uses
- Recover printed text from a clear local image.
- Select the matching English, Simplified Chinese or Japanese model and inspect OCR mistakes.
How to use it
- 1.Choose a recognition model, then select one static PNG/JPEG/WebP.
- 2.Start recognition; engine/model loading and recognition show progress and can be cancelled.
- 3.Review the complete text and save TXT or copy it. New image, language, locale, clear or leaving the tool discards old work/results.
Limits and notes
- One10 MiB image,4096px per edge/4,000,000 pixels; header dimensions checked before native decoding. Local white-background PNG preparation ≤20 MiB; text ≤100,000 UTF-16 units. File reading, native decoding and PNG preparation each have30s deadlines; the complete model phase60s, recognition120s. Cancel terminates the OCR Worker. Limits do not cap all native decoder/WASM process memory. Static PNG/JPEG/WebP only; SVG, GIF, BMP, HEIC/AVIF and animated PNG/WebP are rejected. No input persistence or model IndexedDB cache; normal HTTP asset caching may apply.
- Review every character before reuse. OCR can misread numbers, punctuation, CJK and reading order; handwriting, vertical Japanese, complex tables and page layout are outside this tool’s tested scope. Repeated text is preserved. Preview escapes hidden control/directional characters as\uXXXX; TXT retains the recognized text bytes as UTF-8, without a BOM. No accuracy or legal/archival suitability guarantee.
- The local OCR engine or language model failed to load. Check the site assets/connection and retry. No remote OCR fallback.
Frequently asked questions
Does this upload my image?
No. Only fixed engine/model assets load from this site. Recognition runs in a local Worker; no remote OCR or model IndexedDB persistence.
What if a model cannot load?
The local OCR engine or language model failed to load. Check the site assets/connection and retry. No remote OCR fallback.
Is the recognized text guaranteed correct?
Review every character before reuse. OCR can misread numbers, punctuation, CJK and reading order; handwriting, vertical Japanese, complex tables and page layout are outside this tool’s tested scope. Repeated text is preserved. Preview escapes hidden control/directional characters as\uXXXX; TXT retains the recognized text bytes as UTF-8, without a BOM. No accuracy or legal/archival suitability guarantee.