A lightweight OCR application developed and released by the National Diet Library of Japan. Feed it scanned images of books and magazines, and it recognizes both vertical and horizontal Japanese text and converts them into text data. Its biggest selling point is that no GPU is required — it runs at a reasonable speed on an ordinary laptop. Both a desktop app that works entirely with mouse operations and a command-line version for automating bulk processing are available, and it is released under a CC BY 4.0 license. From organizing personal documents to research and business use, it is free to use.
Key Features
- Runs locally without a GPU: It combines three modules — layout recognition, text-line recognition, and reading-order estimation — using lightweight models, so it runs on a CPU-only PC. On machines with onnxruntime-gpu installed, there is also a beta option to use a GPU via
--device cuda - Supports both vertical and horizontal Japanese text: Designed with modern printed materials in mind, it handles vertically typeset books and magazines as they are. An
--enable-tcyoption is also provided to improve recognition of horizontal-in-vertical mixed text - Experimental support for handwriting and English: Model updates in the v1.2 series improved recognition of Japanese handwriting and English typewriter text. Evaluation results on a handwriting dataset are published in the GitHub README
- Five output formats: TXT, JSON, XML, TEI, and PDF with an invisible text layer. TEI is a format used in digital humanities text research, and the transparent-text PDF makes documents full-text searchable while preserving their appearance
- Available as both an app and a CLI: The GUI app only requires installing it and pointing it at an image folder. If Python 3.10 or later is installed, the command-line version is also available, with
--sourcedirfor batch-processing a folder,--sourceimgfor a single file, and--viz Trueto output visualizations of the recognized regions - Usable in closed environments: Because everything runs locally, scanned images are never sent externally. That makes it easy to use for digitizing confidential documents or paper records containing personal information
Pricing
| Plan | Price | Main features |
|---|---|---|
| Free (CC BY 4.0) | $0 | All features available. No limits on either the app or the CLI version |
Pricing information is current as of August 2026. Please check the official GitHub repository for the latest information. Note that while the tool itself is CC BY 4.0, its dependencies are covered by their own separate licenses.
Pros & Cons
✅ Pros
- Completely free, and usable freely — including commercially — under the terms of CC BY 4.0
- Runs on ordinary laptops without a GPU, so the barrier to getting started is low
- Images are never sent externally, so confidential documents and materials containing personal information can be processed with confidence
- Strong in an area general-purpose OCR struggles with: older vertically typeset books and magazines
- Outputs directly into formats useful for research and archiving, such as TEI and transparent-text PDF
⚠️ Cons
- Handwriting support is positioned as experimental, so it cannot be expected to match the accuracy achieved on printed type
- Structuring tables and complex layouts (such as cell-level extraction) is not its main purpose
- Placing the app in a path containing full-width Japanese characters can cause startup to fail
- At least 1 GB of free memory is required to launch, so it will not run on extremely underpowered machines
- There is no support desk of the kind a commercial service provides; self-service troubleshooting is the basic assumption
Comparison with Similar Services
| Comparison | NDLOCR-Lite | NDLOCR (full version) | YomiToku | Google Cloud Vision OCR |
|---|---|---|---|---|
| Provider | National Diet Library | National Diet Library | Individual developer (OSS) | |
| Environment | CPU only is fine | GPU recommended | GPU recommended | Cloud (API) |
| License | CC BY 4.0 | CC BY 4.0 | CC BY-NC-SA 4.0 | Commercial service |
| Price | Free | Free | Free for non-commercial use | Pay-as-you-go |
| Strengths | Modern printed and vertical text | Large-scale digitization | Japanese documents with layout analysis | Multilingual, general purpose |
| Sends data externally | No | No | No | Yes |
Who Is It For
- People who want to turn printed books, magazines, and booklets on hand into searchable text
- People who work with vertically typeset Japanese materials and were not satisfied with the accuracy of general-purpose OCR
- Organizations that do not want to send images to a cloud OCR service because the materials are confidential or contain personal information
- Libraries, archives, and local history museums that want to advance digitization without a budget
- Developers and researchers who want to batch-process large numbers of images from the command line and feed the results into downstream processing
Summary
NDLOCR-Lite packages the expertise the National Diet Library has built up over years of digitizing materials into a free OCR tool that runs on a PC without a GPU. It is strong on modern Japanese printed materials — especially vertically typeset books and magazines — and offers practical output formats such as TEI and transparent-text PDF. It has limits with handwriting and complex tables, but for the goal of “safely turning the documents on hand into text locally,” it is an option worth trying first. Since a GUI app version is provided, start by trying a few images in the app, and move to the command-line version once the volume grows.