Third-party Models, Libraries and Licenses

Model names, open-source licenses and source downloads for the browser-local video and GIF tools.

Last updated: 2026-11-02

The browser tools use the models and libraries below. Their original license terms and copyright notices apply. These notices do not transfer ownership of your selected media or guarantee a particular processing result.

Portrait models

Fast processing

MediaPipe SelfieSegmenter, square float16 v1 — Google / MediaPipe Authors

Apache License 2.0, as stated in the official model card. The asset manifest records the source and SHA-256.

Detailed edges

MODNet photographic portrait matting, official ONNX weights — Zhanghan Ke and contributors

Apache License 2.0. The upstream project licenses its code, models and demos under Apache 2.0, excluding its documentation GIFs. Our asset manifest records the selected model hash.

The model weights are unmodified. Portrait inference runs on your device. Color-key video and GIF removal do not use a portrait model. Model limits and browser recommendations are described in Help.

Processing libraries

MediaPipe Tasks Vision 1.0.1 — Fast portrait inference. Apache 2.0 · Google / MediaPipe source
ONNX Runtime Web 1.22.0 — Detailed-edge inference. MIT · Microsoft release source
MediaBunny 1.61.0, with CutoutUp diagnostic changes — Media reading, encoding and export. Mozilla Public License 2.0 · Upstream source · Download the corresponding modified source
libopus-wasm 0.4.1 — Local fallback for Opus audio encoding. MIT · Wrapper source
libopus 1.6.1, included in that wrapper — Opus audio codec. BSD-style license · Xiph.Org release source
gifuct-js 2.1.2, gifenc 1.0.3 and js-binary-schema-parser 2.0.3 — Animated GIF decoding and encoding. MIT license texts, authors and source links

MediaBunny source changes

CutoutUp adds fixed diagnostic headings to codec error stacks so browser failures can be classified consistently. The changes cover the source files and published JavaScript entries. They preserve the original error, callback flow and codec settings; they do not claim to fix a codec failure. The source download includes the library source with these changes, the patch, package metadata and MPL 2.0 license. The original license notices are retained.

Demonstration media

Current portrait and product clips are disclosed as AI-generated; some GIF examples are original graphic animations. They are samples of the actual tools, with no manual subject retouching. Case cards link their own records; see the portrait source manifest, tool sample manifest and sample-source explanation. These are separate from the model and library licenses above.