OCR Text Recognition

Drag & drop an image here, or click to select a file

Supports PNG / JPG / JPEG / BMP / WebP, single file ≤ 10MB

The OCR engine is powered by PP-OCRv6 (ppu-paddle-ocr). All recognition happens locally in the browser — images are never uploaded to a server. WebGPU acceleration is supported.

Online OCR based on PP-OCRv6. Converts Chinese/English images to text locally — images are never uploaded.

How to Use & FAQ5 steps · 8 Q&A

OCR Text Recognition

Online OCR based on PP-OCRv6. Converts Chinese/English images to text locally — images are never uploaded.

How to Use

  1. Upload or drag in an imageSupports PNG / JPG / JPEG / BMP / WebP, up to 10MB; clear and upright images work best.
  2. Choose a recognition languageFor mixed layouts pick "mixed"; for a single language pick that option.
  3. Click Start recognitionRecognition completes locally; the first run loads the model first, please wait.
  4. Proofread and copyRead through the result, fix key numbers/names, then copy or download .txt.
  5. Tips to improve accuracyZoom the screenshot, crop irrelevant areas, raise contrast, or split a long image and stitch results.

FAQ

How accurate is the OCR?

Based on the PP-OCRv6 model, Chinese and English recognition works well. Accuracy depends on image clarity, shooting angle and font; use clear, upright images. Handwriting, decorative fonts, heavy skew or low contrast lower the rate.

Will the image be uploaded to a server?

No. Recognition runs entirely in the browser (with WebGPU acceleration) — the image never leaves your device, so it is privacy-safe; you can safely recognize screenshots with sensitive info.

Which recognition languages are supported?

It supports mixed Chinese-English, Chinese-only and English-only modes, switchable as needed. For mixed-layout documents, "mixed" is usually the most stable.

Why is the first recognition slower?

The first run needs to download and initialize the OCR model (a large file); afterwards it is reused and later runs are much faster. The page preloads the model to shorten the wait.

Which image formats and sizes are supported?

PNG / JPG / JPEG / BMP / WebP are supported, up to 10MB per file. Too large or too small both hurt quality; crop irrelevant margins before recognizing.

Can I export the recognition result?

Yes. After recognition you can copy with one click or download it as a .txt file, and the history is stored locally in your browser for later review.

What if the recognition is wrong?

Try: raise resolution, straighten the angle, boost contrast, or zoom in and recognize in segments. For proper nouns or numbers, always manually verify the key fields afterwards.

Can it recognize PDFs or tables directly?

It mainly accepts images for now. Export a PDF to images (or screenshot) first; tables are output line by line, so complex row/column structure needs manual cleanup afterwards.