NOTICE

Clips Studio
Copyright (C) 2026 ColinGPT9

This program is free software: you can redistribute it and/or modify it
under the terms of the GNU Affero General Public License as published by the
Free Software Foundation, either version 3 of the License, or (at your
option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU Affero General Public License
for more details.

You should have received a copy of the GNU Affero General Public License
along with this program. If not, see <https://www.gnu.org/licenses/>.

The full text is in LICENSE.


-------------------------------------------------------------------------------
What this means if you are just using the app
-------------------------------------------------------------------------------

Nothing changes for you. Download it, install it, clip your streams, post the
clips, make money from them. The AGPL places obligations on people who
DISTRIBUTE the software or run a MODIFIED version as a network service — not
on people who use it. Your videos and your clips are yours; the licence has
nothing to say about them.

If you modify Clips Studio and give that modified version to someone else, or
run it as a service other people use over a network, you have to offer them
the source of your modified version under this same licence.


-------------------------------------------------------------------------------
Parts under other licences
-------------------------------------------------------------------------------

The plugin SDK in sdk/python/ (clipskitty_sdk, which the installer also
bundles so plugins can import it) is MIT-licensed; its notice is in
sdk/python/LICENSE and travels with it. The example plugins in
examples/pipelines/scene-cut-highlights/ and examples/pipelines/keyword-rater/
are MIT as well, and the copy of the Awesome Clips Kitty list in
plugins/catalog_index.json is CC0-1.0, as is the list itself
(github.com/ColinGPT9/awesome-clips-kitty). Plugins, pipelines and apps made by other people keep their own
licences; installing one does not change it, and nothing here changes theirs.

-------------------------------------------------------------------------------
Why a copyleft licence
-------------------------------------------------------------------------------

It keeps improvements to the tool available to the creators it was written
for, rather than allowing a closed fork of the same work to be sold back to
them.


-------------------------------------------------------------------------------
Third-party components
-------------------------------------------------------------------------------

The Windows installer bundles the Python engine and its dependencies. The
significant ones, with their licences:

  ultralytics (YOLOv8)      AGPL-3.0     pose/face tracking + model weights
  TalkNet-ASD               MIT          active speaker detection; the model
                                         code is vendored unmodified under
                                         third_party/talknet/ (c) 2021 Tao
                                         Ruijie, and its pretrained weights
                                         are fetched by scripts/fetch_asd_model.py
  PANNs (Qiuqiang Kong)     MIT          game sounds; the MobileNetV1 network
                                         is adapted in analysis/panns.py
                                         (c) 2018-2020 Qiuqiang Kong, and its
                                         pretrained weights (Zenodo 3987831)
                                         are CC BY 4.0, fetched by
                                         scripts/fetch_panns.py. AudioSet's
                                         class names (config/
                                         audioset_labels.txt) are CC BY 4.0,
                                         Google
  E-BARD detection models   CC BY 4.0    the basketball, the hoop and the
                                         players, for basketball's 9:16 framing
                                         (sports/basketball/crop/sources.py);
                                         the YOLOv8n weights (huggingface.co/
                                         GabrieleGiudici/E-BARD-detection-
                                         models, revision 3f4789c), unmodified,
                                         fetched by scripts/
                                         fetch_basketball_model.py. Trained
                                         with ultralytics, so its AGPL-3.0
                                         terms apply as to the YOLOv8 weights;
                                         full attribution below
  pyannote segmentation-3.0 MIT          who is talking when, for the second
                                         speaker's caption colour (analysis/
                                         voice_turns.py); (c) 2022 CNRS, the
                                         ONNX export is sherpa-onnx's (k2-fsa),
                                         fetched by scripts/fetch_voice_model.py
  WeSpeaker ResNet34-LM     CC BY 4.0    whose voice a turn is, for the same;
                                         pretrained on VoxCeleb by the
                                         WeSpeaker team (huggingface.co/
                                         Wespeaker/wespeaker-voxceleb-
                                         resnet34-LM), unmodified, fetched by
                                         scripts/fetch_voice_model.py
  RapidOCR                  Apache-2.0   on-screen text in a gaming stream;
                                         its bundled PaddleOCR models are
                                         Apache-2.0 (PaddlePaddle)
  faster-whisper            MIT          transcription; its model weights are
                                         fetched by scripts/fetch_whisper.py
                                         and are MIT (Systran/faster-whisper,
                                         converted from OpenAI Whisper)
  keyring                   MIT          OS-protected Connected Accounts tokens
                                         on Linux and macOS only; Windows uses
                                         DPAPI, and keyring isn't installed there
  websockets                BSD-3-Clause desktop Twitch EventSub connections
  Ollama                    MIT          the local AI runtime, bundled
                                         unmodified and fetched at build time
                                         by scripts/fetch_ollama.py
  PyTorch                   BSD-3-Clause tensor runtime, CUDA libraries
  OpenCV (opencv-python)    Apache-2.0   frame decoding and analysis
  yt-dlp                    Unlicense    downloading source videos
  yt-dlp-ejs                Unlicense,   YouTube player solver scripts, for the
                            MIT, ISC     YouTube cookies fallback only
  QuickJS-ng                MIT          runs those scripts; bundled unmodified
                                         and fetched at build time by
                                         scripts/fetch_quickjs.py
  FFmpeg                    see below    all encoding, decoding and muxing
  FastAPI / Uvicorn         MIT          the local API the desktop app talks to
  Electron                  MIT          the desktop shell
  React                     MIT          the interface
  Material Icons            Apache-2.0   a few icon shapes in the layout
                                         editor's platform preview (Google)
  Lucide                    ISC          a few icon shapes in the same preview;
                                         both notices are in
                                         third_party/icons/NOTICE.md

The E-BARD detection model (https://github.com/GabrieleGiudic/E-BARD) is licensed under
Creative Commons Attribution 4.0 International (CC BY 4.0,
https://creativecommons.org/licenses/by/4.0/). Clips Kitty uses its YOLOv8n
weights unmodified:
  model    https://huggingface.co/GabrieleGiudici/E-BARD-detection-models
  code     https://github.com/GabrieleGiudic/E-BARD
  dataset  https://huggingface.co/datasets/GabrieleGiudici/E-BARD-detection
E-BARD extends BARD: G. Giudici, A. Maurino and P. Zuccolotto, "BARD: A
Basketball Action Recognition Dataset for multi-label classification",
Computer Vision and Image Understanding 267 (2026) 104713,
https://doi.org/10.1016/j.cviu.2026.104713.

FFmpeg is fetched at build time by scripts/fetch_ffmpeg.py rather than kept
in this repository. The builds used are LGPL-2.1-or-later, or GPL-2.0-or-later
where they include GPL-licensed components; the exact terms ship with the
binary in the installer.

Ollama is fetched at build time by scripts/fetch_ollama.py rather than kept in
this repository, and ships unmodified. Its MIT licence travels with it in the
installer as README-OLLAMA.txt.

The language models Ollama runs are NOT bundled. The app downloads the one it
needs on first launch, and each carries its own licence — Gemma, for example,
is covered by the Gemma Terms of Use rather than by anything here. That is
also why they are downloaded rather than shipped: redistributing them would
mean accepting those terms on every user's behalf.


-------------------------------------------------------------------------------
Commercial licensing
-------------------------------------------------------------------------------

If the AGPL does not suit your use, ask. The copyright is held by one person,
which means a different arrangement is possible without collecting agreements
from a hundred contributors.