VideoAnnotator runs computer-vision and audio pipelines over videos of people (faces, bodies, speech, scenes, and labels from a vision-language model), keeps a record of what made every result, and shows the results on the video in its viewer. It runs on your own computer: videos never leave it.
New here? Start with Getting started.
- Installation: Windows, macOS and Linux, with or without a GPU, and the dev container.
- Docker: the CPU and GPU images, and Docker Compose.
- Access tokens for gated models: the Hugging Face token that speaker diarization needs.
- Installation troubleshooting
- Getting started: start the server, open the viewer, run a first job.
- Command-line examples
- Pipelines: what each pipeline does and the files it writes.
- Getting your results: downloading and reading the output files.
- Configuration and environment variables
- Connecting the viewer with an API key
- Scene detection and person tracking
- Troubleshooting
- Contributing guide
- Development notes: conventions, output formats, the pre-commit hooks.
- Testing overview, testing standards and coverage
- Roadmap for v1.6.0 (consolidation), then v1.7.0 (public release), v1.8.0 (plugins) and v1.8 to v2.0
- Changelog
Older documents, kept for the record, are in archive/. They describe earlier versions and may be wrong about this one.