Short-form clipping is not just a highlight-detection problem: the recurring cost and lock-in of hosted editing pipelines can matter as much as the AI itself. This project’s key tradeoff is to make the workflow inspectable and self-hostable, shifting control to the operator while also shifting operational work and provider costs there.
What Sets It Apart
- Segment selection is based on a transcript, with LLM providers ranging from hosted models to Ollama. That makes the selection layer replaceable rather than tied to a single model vendor.
- The pipeline combines selection with face-aware vertical framing, timed captions, and optional B-roll, reducing the handoffs between finding a moment and producing a platform-ready clip.
- Self-hosting under AGPL-3.0 offers source-level customization and avoids dependence on a hosted editing workflow; it does not make transcription, compute, or every model provider free.
- Publishing integrations and performance feedback connect clip production to distribution, although they also make account permissions and platform-specific behavior part of the setup and upkeep.
Who It’s For
Great fit for creators or teams that process videos regularly and value control over their content, model choices, and processing limits. Look elsewhere if you want a fully managed service with no infrastructure work, or if you need guaranteed clip quality without reviewing AI-selected moments. Self-hosting still requires technical upkeep, and transcription or hosted model usage may incur separate costs.
