Highlights
- Adds browser-local semantic optical-flow tracking with cohort trajectories, adjustable sampling, vector density, trail persistence, localized progress, and reusable rendered results.
- Mirrors every voice model used by Timeline Studio to owned Hugging Face and ModelScope repositories, pinned to immutable revisions.
- Routes Chinese sessions through ModelScope first with automatic Hugging Face fallback and a shared provider-independent cache identity.
- Improves Effects and AI Voice behavior across desktop and the 412×915 mobile layout.
- Updates Timeline Studio to version 0.8.3.
Validation
- ESLint: 0 errors (existing warnings remain)
- TypeScript: passed
- Production Vite build: passed
- Desktop and 412×915 mobile AI Voice panels verified
- Chinese Piper voice generation verified end to end
Model licensing
The owned voice-model mirror preserves upstream source notes, checksums, and individual licenses. Some included voice models are restricted to non-commercial use; review the mirror model card before redistribution or commercial deployment.