Licences
The first time you use an AI tool, it asks before downloading the software and models it needs, and then runs them on your device. These are their licences. Each part links a text file with the full licence texts and notices.
ONNX Runtime Web 1.30.0
Runs the face and background models. Its JavaScript is in the AI worker, and its WebAssembly is downloaded. MIT licence, Copyright (c) Microsoft Corporation. The WebAssembly also holds code from other projects, among them ONNX, Abseil and FlatBuffers (Apache 2.0), Protocol Buffers and RE2 (BSD 3-Clause), Microsoft GSL, SafeInt and JSON for Modern C++ (MIT), Eigen (Mozilla Public License 2.0) and Emscripten (MIT). ONNX Runtime's third-party notices, in the file below, list them all with their licences. Eigen's source code is at gitlab.com/libeigen/eigen, ONNX Runtime's at github.com/microsoft/onnxruntime.
ONNX Runtime's licence and third-party notices
Tesseract.js 7.0.0
Reads text. Its JavaScript is in the AI worker, and its worker script is downloaded. Apache License 2.0. It bundles regenerator-runtime, bmp-js, is-url, zlib.js and buffer (MIT), ieee754 (BSD 3-Clause), and wasm-feature-detect and idb-keyval (Apache 2.0).
Tesseract.js's licence and those of what it bundles
tesseract.js-core 7.0.0
Tesseract itself, compiled to WebAssembly with Emscripten, and downloaded. Apache License 2.0. It includes Tesseract (Apache 2.0); Leptonica (Copyright (C) 2001-2020 Leptonica); libjpeg from the Independent JPEG Group (Copyright (C) 1991-2014, Thomas G. Lane, Guido Vollbeding); libpng (Copyright (c) 1995-2020 The PNG Reference Library Authors); libtiff (Copyright (c) 1988-1997 Sam Leffler, Copyright (c) 1991-1997 Silicon Graphics, Inc.); libwebp (Copyright (c) 2010, Google Inc.); giflib (Copyright (c) 1997 Eric S. Raymond); zlib (Copyright (C) 1995-2022 Jean-loup Gailly and Mark Adler); openlibm; and Emscripten's runtime (Copyright (c) 2010-2014 Emscripten authors).
This software is based in part on the work of the Independent JPEG Group.
The licences of tesseract.js-core and everything it includes
Models
- The English and German text models, eng.traineddata and deu.traineddata from tessdata_fast 4.1.0: Apache License 2.0. Each downloads on its own, the first time text is read in that language.
- YuNet, the face model from OpenCV Zoo: MIT licence, Copyright (c) 2020 Shiqi Yu. Changed for Edit Image: its input size is free rather than 640 × 640, and the shapes it recorded for that size are removed. The weights are unchanged.
- BiRefNet-lite, the background model from BiRefNet: MIT licence, Copyright (c) 2024 ZhengPeng. Its backbone, Swin Transformer, is Microsoft's code and started from Microsoft's ImageNet-22k weights: MIT licence, Copyright (c) Microsoft Corporation. Changed for Edit Image: its input size is free and only its final mask is kept, its deformable convolutions are written as plain sampling and convolutions with the same weights, the weights of its convolutions and linear layers are stored as 8-bit numbers (3.25 MB of smaller values stay as they were), and the exporter's notes are removed.