ImgPeel
ImgPeel · Peel OCR

Guide

ImgPeel (Peel OCR) is an everyday OCR tool: drop, pick, or paste an image to detect and recognize text in your browser. Unlike most OCR sites, your images are never sent to a server — after model weights download, detection and recognition run entirely on-device, and image data stays in this tab.

The engine is PaddleOCR’s PP-OCRv6. The browser prefers WebGPU, falling back to WebGL and WebAssembly. Chinese–English mixed text is supported. No sign-up, no usage caps, no watermarks.

How to use

  1. Choose images

    Drop into the upload area, click to select (multi-select OK), or paste a screenshot with Ctrl/⌘V. Up to 20 at a time.

  2. Recognize on-device

    The model processes the queue in order with live progress. Images never leave your device.

  3. Copy or export

    Review lines and confidence, copy all text, or export .txt / .md / .docx. For batches, zip packs one file per image.

Choosing a model

Accuracy and download size trade off, so three weight packs are available — switch anytime. Downloaded models stay cached locally.

Fast

PP-OCRv6 TinyAbout 6 MB

Smallest size and fastest load — enough for screenshots and clean print docs; best default on mobile networks.

Balanced

PP-OCRv6 SmallAbout 30 MB

Fuller dictionary and steadier on small type and complex layouts — recommended for everyday desktop use.

High accuracy

PP-OCRv6 MediumAbout 132 MB

Best accuracy, at the cost of 100+ MB weights and higher memory — use on desktop over Wi‑Fi.

FAQ

Are my images uploaded to a server?
No. After models download to the browser, detection and recognition run on your CPU/GPU. Image data stays in this tab’s memory and never hits a backend. You can disconnect the network while recognizing to verify.
Which languages and image formats are supported?
Simplified Chinese, English, digits, and common symbols — mixed Chinese–English needs no language switch. Images: PNG, JPG, WebP via drag-and-drop, file picker, or Ctrl/⌘V paste.
How do I choose among the three models?
Fast is smallest and quickest — good for screenshots and clear docs. Balanced trades size for a fuller dictionary. High accuracy is best but downloads 100+ MB — prefer Wi‑Fi on desktop.
Why does the first visit take a while?
ONNX weights must download once. After that they are cached locally and reopen without re-downloading.
Can I use it on a phone?
Yes — the UI works on phones and tablets; you can shoot or pick from the album. High accuracy is heavy on mobile networks and memory; prefer Fast or Balanced on phones.
How do I export results?
Lines list with confidence, color-matched to boxes on the image. Copy all text, or download .txt / .md / .docx for the current image. After a batch, Zip txt / md / docx packs one file per image.
Do I need an account or payment?
No. Open the page and use it — no login, quotas, or watermarks.
What if recognition is inaccurate?
Prefer sharp, upright text with good contrast. Handwriting, stylized fonts, heavy compression, and low resolution hurt accuracy — try a higher model tier.

Roadmap

  • More interface languages