Repository navigation
Kiteva — AI Motion Graphics Agent now lives at delexw/motion-graphics-agent. This release brings together the production workflow improvements since v0.4.0.
Added
- Read-only Production Desk with timeline inspection, source audio waveforms and inline storyboard viewing.
- Ming 3.0 TTS 0.5B narration with neutral voices, supported-language selections and original narrator cloning.
- Shared visuals with language-specific narration and subtitle tracks, plus cue-level subtitle progress.
- Refinement history, attachments and question/answer continuity, including cancelled versions and monotonically increasing version numbers.
Changed
- Separate music and SFX source files with timed placements; the agent retains ownership of the final music/SFX mix.
- Centralized workspace paths separate render artifacts and assets from logs, diagnostics and runtime data.
- Bounded factual cache evidence replaces Haiku change summaries; deterministic checks handle provable cache decisions.
- Structured correction outcomes escalate unsuccessful Haiku repairs to the main model with context.
- Activity history loads recursively on demand and releases collapsed branches; playback uses the media clock and avoids competing audible players.
Fixed
- Brief recovery and draft removal, shared attachment previews, normalized download names and consistent project controls.
- Duration validation accepts the target length within five seconds; renderers retain their native execution defaults.
- Shared credential boundaries and test guards prevent codebase tests from using real provider credentials or external APIs.
This is a GitHub source release, without desktop installers, npm publication, bundled local models or downloaded upstream skills. Windows support remains experimental. Existing Qwen speech selections are normalized to the retained narration providers; both Qwen providers have been retired.