NOTICE
Rama main · réplica de lectura del repositorio original.
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.