Dublia 1.0.0 — third-party notices

Dublia is developed by DRS LLC — Digital Research & Development.

Bundled runtime components
--------------------------
The portable build includes a Nuitka standalone Python 3.11 runtime, Tk/Tcl
runtime data, application modules, and a yt-dlp downloader. Their respective
licenses remain in effect. Python is distributed under the Python Software
Foundation License; Tcl/Tk and yt-dlp have their own licenses. This notice does
not replace the license texts of those projects.

FFmpeg
------
The portable test build includes FFmpeg and ffprobe 9.0.2, "essentials_build"
from https://www.gyan.dev/ffmpeg/builds/. The reported build configuration
includes --enable-gpl and --enable-version3. FFmpeg is licensed under the GNU
General Public License (GPL) version 3 for this build.

Redistributing this GPL-enabled FFmpeg build with a commercial application
creates source-availability and notice obligations. Before commercial
distribution, DRS LLC must either comply with those obligations (including
providing the corresponding source as required) or replace the bundled FFmpeg
with a build whose license is compatible with the distribution plan. This test
package must not be treated as a completed legal review.

FFmpeg project: https://ffmpeg.org/
FFmpeg license information: https://ffmpeg.org/legal.html
Gyan.dev builds: https://www.gyan.dev/ffmpeg/builds/

FaceFusion lip sync
-------------------
The setup wizard downloads FaceFusion 3.9.2 from the official GitHub project on
request and installs an isolated Python runtime for it. FaceFusion is licensed
under the OpenRAIL-AS license; review its use restrictions before commercial
distribution.

The Dublia lip-sync path uses the `edtalk_256` model, which the FaceFusion
source identifies as Apache-2.0. It does not use the `wav2lip` models, which
are marked non-commercial in the FaceFusion model metadata. Other FaceFusion
face-detection, landmark, parser, recognizer and content-analysis models have
their own licenses and must be reviewed before redistribution. Dublia downloads
them from FaceFusion's official model assets at setup time; they are not bundled
in the installer.

FaceFusion project: https://github.com/facefusion/facefusion
FaceFusion license: https://github.com/facefusion/facefusion/blob/master/LICENSE.md

NVIDIA CUDA and cuDNN
---------------------
On PCs with a compatible NVIDIA GPU, setup can install CUDA runtime, cuBLAS,
cuFFT, cuRAND, NVRTC, NVJITLink, cuDNN and GPU ONNX Runtime packages from PyPI
into the isolated FaceFusion runtime. These are NVIDIA CUDA/cuDNN redistributable
components and are subject to the NVIDIA CUDA Toolkit and cuDNN license terms.
They are downloaded only after setup consent and are not bundled in the Dublia
installer. No NVIDIA API key is required for this local runtime use.

NVIDIA CUDA toolkit: https://docs.nvidia.com/cuda/eula/
NVIDIA cuDNN: https://docs.nvidia.com/deeplearning/cudnn/legal-notice/
ONNX Runtime: https://github.com/microsoft/onnxruntime

Embedded Python
---------------
The FaceFusion runtime uses the official Python 3.12 embeddable distribution
from python.org, downloaded at setup time. Python is distributed under the
Python Software Foundation License.

Downloaded with setup consent (not bundled)
-------------------------------------------
- Ollama and Ollama model weights: separate installation and licenses.
- faster-whisper / CTranslate2 / Hugging Face models: separate licenses and
  model terms; model weights are downloaded separately, not bundled.
- Edge TTS: network service subject to its own terms (female default).
- FaceFusion 3.9.2 runtime and required models: downloaded from the official
  FaceFusion project after setup consent; not bundled in the installer. See the
  FaceFusion section above for the OpenRAIL-AS and model-license review.
- Chatterbox-TTS 0.1.7 and PyTorch/Torchaudio: downloaded from PyPI by setup
  into an isolated runtime; not bundled. Chatterbox is MIT-licensed.
- Chatterbox multilingual model (ResembleAI/chatterbox) and the Spanish
  fine-tune (ResembleAI/Chatterbox-Multilingual-es-es): downloaded from
  Hugging Face; both model cards list MIT. Review the current model cards and
  included notices before distribution.

Optional external components (not included)
-------------------------------------------
- pyannote.audio and PyTorch diarization stack: not included in this baseline
  engine; optional future diarization build would require separate license
  review and a Hugging Face token for gated models.

No payment credentials, private signing keys, API keys, or user data are
included in this package.
