No need to save a file first; copy a screenshot and paste it anywhere on the page.
Images stay local. Text comes with you
Turn visible words
into editable text.
Paste a screenshot or add images and scanned PDFs. Extract English, Chinese, and numbers, then copy or download the result.
LOCAL OCR
Image & PDF OCR
Files are never uploaded
INPUT PAGES
Pages to Process
No files added
＋Waiting for images or PDFs
You can add multiple files in sequence.
Renders each page and prefers existing selectable PDF text when available.
The OCR engine and language models run in your browser; files are never sent to a server.
Check whether you need OCR before compressing
Try selecting text in your source PDF first. If it is already searchable, preserve that source. For scans, use the practice page to check the reference code, the amount 123.45, and similar-looking O/0 characters. OCR output needs comparison with the original.
Read the example, samples, and exact rules →Prepare the source before trusting extracted text
OCR turns visible letter shapes into editable text, but it is not a guarantee of accuracy. Clean alignment, sufficient resolution, the correct language, and a final human review matter more than simply running the recognizer twice.
Step by step
- Start with the clearest source
Use the original scan or photo when possible. Crop unrelated borders, straighten the page, avoid shadows, and make sure small characters are not blurred.
- Select the document language
Choose the language that matches the page. Mixed-language pages can take longer and may confuse similar Latin, Chinese, numeric, or punctuation characters.
- Review page by page
Check names, dates, amounts, decimal separators, account numbers, tables, and line breaks against the source instead of accepting a long block at once.
- Export a working copy
Download plain text for editing and keep the source image or PDF beside it. Treat OCR output as a draft when accuracy has legal, medical, academic, or financial consequences.
Decision guide
- Clean scan
- Usually produces the most reliable result when text is level, high contrast, and large enough.
- Phone photo
- Works best with even lighting, a parallel camera angle, and no fingers or page curvature.
- Scanned PDF
- Each page must be rendered before recognition, so long documents take more time and memory.
- Handwriting and tables
- These are harder than printed paragraphs and often require manual correction or a specialist workflow.
Important limits
- OCR can confuse visually similar characters such as O and 0, l and 1, or punctuation marks.
- Complex tables, columns, stamps, handwriting, and low-contrast backgrounds can lose reading order.
- Recognition quality depends on the source; enlarging a blurred image does not recreate missing detail.
- Always compare consequential values with the original document before using or sharing the text.
04 · Common questions
Common questions
Why are names and numbers often wrong?
OCR predicts characters from their shapes and context. Proper nouns, codes, and isolated numbers provide less context and need closer review.
Can OCR recover text from a very blurry photo?
It may recover fragments, but sharpening and enlargement cannot restore detail that was never captured. A new scan or photo is preferable.
Are documents uploaded?
No. Recognition runs with browser resources on the current device; the selected document is not sent to OmniKit.
Private OCR for images and scanned PDFs
Paste screenshots or add PNG, JPG, WebP, BMP, and PDF files. OmniKit reads existing selectable PDF text first and runs OCR only where needed.
The OCR engine and language data run locally in your browser. Review confidence, edit the extracted text, then copy it or download a TXT file.