You can’t judge an OCR engine from a datasheet.

Accuracy depends on your labels, your lighting, your print quality and your character set — not on a benchmark someone else ran. The only honest test is to put your image through the engine and look at the result.

Web OCR is a free tool that does exactly that — in your browser. Drop in an image and it reads the text on the spot. Nothing to install, no account, no upload. The whole pipeline runs on your machine.

Open the free Web OCR tool →

What it does

• Runs entirely in the browser — the model loads, the image is analysed and the text is recognised on your own device. There is no backend, no Python, no model server.
• Nothing is uploaded — your image never leaves the browser. That matters when the label carries a serial number, a batch code or anything commercially sensitive.
• Choose speed or accuracy — switch between model tiers and see the difference immediately.
• Annotated results — recognised text is drawn back onto the image with bounding boxes, so you can see exactly what was read and where.

How it works

Layer	Technology
OCR engine	PP-OCRv6 (tiny / small / medium)
Inference runtime	ONNX Runtime Web
Acceleration	WebGPU (WebGL fallback)
Image processing	OpenCV.js (WASM)
Build	Vite

Everything is standard, client-side technology. Load the page, the model is fetched once and cached, and recognition takes roughly 0.2–0.5 s for the lighter tiers on a typical machine.

Pick your trade-off: four model tiers

PP-OCRv6 ships in several sizes, and the tool lets you switch between them:

Tier	Model size	Best for
PP-OCRv5	~25 MB	Legacy comparison
PP-OCRv6 tiny	~6 MB	Speed — runs well even without a strong GPU
PP-OCRv6 small	~30 MB	The everyday balance of speed and accuracy
PP-OCRv6 medium	~138 MB	Maximum accuracy — the largest tier

The newest addition is medium: the high-accuracy tier, roughly four times the parameters of small, tuned for the hardest captures. It is heavier to download and runs best on a machine with a decent GPU — but when a label is low-contrast, skewed or densely printed, it is the tier that reads it.

The same tiers are what ship in our products: tiny for speed on the line, small for the everyday balance, and medium where accuracy is everything.

What to test it with

Bring your own worst cases — that is where a tool like this earns its keep:

• Printed label serials and batch codes — the exact strings your system needs to capture.
• Laser-etched or dot-peen marks on metal — hard for conventional OCR.
• Dates, lot numbers and expiry codes on a range of substrates and print qualities.
• Anything you would otherwise type by hand — see whether the engine can read it before you commit to a project.

If the tier you need reads your images, you have a straight answer. If it doesn’t, you have found out for free — and we can talk about what would.

One engine, three product lines

Web OCR is not a toy kept separate from the products — it is the same PP-OCRv6 engine family, exposed so you can test it directly:

• Scanner OCR V2 — turn a barcode scanner into an OCR reader; pull the trigger and the text lands in the active field.
• Fixed-Mount OCR V2 — read characters and grade barcode quality on the line, from one scan.
• Android OCR — recognise text on an Android PDA and type it into the business app.

And all three run on the same shared OCR framework — see How one OCR framework powers three product lines. The browser tool is the simplest way to see what that engine can do before you talk to us.

Requirements and limits

• Browser: a modern Chromium-based browser (Chrome or Edge) for the best performance; WebGPU acceleration where available, WebGL as a fallback.
• First load: the model is downloaded once (6 MB to ~138 MB depending on the tier) and cached afterwards.
• Interface language: Chinese today; the English build ships in October 2026.
• It is a test tool, not a workflow: there is no batch mode, no automation and no integration — those live in the products above and in custom work.

---

💬 Questions or a custom requirement?

This global site is brand new, and our online tools are currently in Chinese. Tell us what you need and we’ll set up a fast, tailored solution for you — an English version is coming in the next few weeks.

📧 sales@happylab.me · 🌐 global.happylab.me

---

HappyLab builds practical AI and browser-based tools for industrial device workflows.