Effective date: September 14, 2026 · Provider: Moves and Measures, LLC
Some of what BrandTactics does runs on your own device, with open-source software and models it sends there from its own address. Each is listed here with its licence.
Remove background runs IS-Net, from “Highly Accurate Dichotomous Image Segmentation” by Xuebin Qin, Hang Dai, Xiaobin Hu, Deng-Ping Fan, Ling Shao and Luc Van Gool (github.com/xuebinqin/DIS), licensed under the Apache License, Version 2.0. BrandTactics converted it to fp16: its weights are stored as half precision and computed as full precision; the network is otherwise unchanged.
A clip’s Remove background — and, on a device with a graphics chip, the frames of a video you make — cut the person out with MODNet, from “MODNet: Trimap-Free Portrait Matting in Real Time” by Zhanghan Ke et al. (github.com/ZHKKKe/MODNet), licensed under the Apache License, Version 2.0. BrandTactics sends the ONNX export published at huggingface.co/Xenova/modnet, its weights stored as half precision, unchanged.
A clip whose cut-out follows the person is drawn, as it plays, through MediaPipe Tasks Vision 1.1.1and its Selfie Segmenter, by Google (github.com/google-ai-edge/mediapipe), licensed under the Apache License, Version 2.0. BrandTactics sends them unmodified.
Captions are heard on your device through Transformers.js 3.8.1, by Hugging Face (github.com/huggingface/transformers.js), licensed under the Apache License, Version 2.0, with the build of ONNX Runtime Web it ships (MIT, below). BrandTactics sends them unmodified.
The speech model is Whisper tiny.en, by OpenAI (github.com/openai/whisper, MIT License; Apache License, Version 2.0 on its model card), as onnx-community’s word-timed export.
A video is read and written on your device (a clip in the Library, its light copy for editing, a video you make) by Mediabunny 1.65.2, by Vanilagy (github.com/Vanilagy/mediabunny), licensed under the Mozilla Public License, Version 2.0. BrandTactics sends its source unmodified, built into its own scripts.
When your browser cannot write a video’s sound as AAC itself, Studio sends it Mediabunny’s AAC encoder extension (@mediabunny/aac-encoder 1.65.2, Mozilla Public License, Version 2.0), which carries the AAC encoder of FFmpeg (ffmpeg.org) built as WebAssembly, licensed under the GNU Lesser General Public License, version 2.1 or later — no GPL, version 3 or non-free parts. It is loaded only while a video is made, and never in a chat’s card. BrandTactics sends it unmodified; its source is FFmpeg at commit 60a2818f76dcd1fee3155b1791304af104a0fc1b(github.com/FFmpeg/FFmpeg/commit/60a2818f76dcd1fee3155b1791304af104a0fc1b) and the extension’s own (github.com/Vanilagy/mediabunny, packages/aac-encoder), whose README gives the exact build; to use a modified FFmpeg, rebuild it with those steps.
ONNX Runtime Web 1.29.0, and the build Transformers.js ships, Copyright (c) Microsoft Corporation (github.com/microsoft/onnxruntime), licensed under the MIT License.