A local-first AI video studio that refuses to fake your timeline

What it does

It’s a desktop-first (Flask + PyWebView, Windows 10/11) video editing workbench that runs the whole pipeline in one place: shot segmentation, AI-generated commentary scripts, TTS voiceover, audio-video sync, subtitles, remixing, and export — including export as a native JianYing (剪映) draft you can open directly in the pro editor. Data lives in local SQLite; rendering goes through FFmpeg.

The interesting bit

The sync engine has an unusually honest design philosophy: if a source shot is too short to cover the voiceover even with safe slow-motion, it explicitly refuses rather than freezing the last frame to fake the duration. Same for target runtime — verified with FFprobe, and if the material can’t reach it, it says so instead of looping clips. That’s rarer in auto-editing tools than it should be.

Key highlights

• Shot segmentation via TransNetV2 with OpenCV histogram fallback, no frame cap
• 12 narration styles, 12 writing personas, 4 narrative perspectives, 10 opening hooks — the style knobs are… thorough
• Music-beat remix mode using librosa with dense/medium/sparse cut densities
• Multiple TTS engines (Edge, Azure, gTTS, Volcano) plus local voice cloning via IndexTTS2 in an isolated Python 3.10 environment
• AI cover generation across many providers, with a local template fallback when clouds are unreachable
• Bilingual Chinese/English UI covering 200+ strings

Caveats

• Windows-only, and it’s a big integration project — cloud LLMs, TTS, and cloning features depend on your own API keys and model weights being in place
• License is restrictive: source is public for personal study and use only; any commercial use requires written authorization from the author
• Voice cloning requires explicit consent from the reference speaker (consent=true is enforced at the interface level)

Verdict

Worth a look if you produce Chinese-style commentary or remix videos on Windows and want an all-local pipeline instead of juggling five tools. Skip it if you need cross-platform support, commercial rights, or a small codebase — this is a large glued-together system, though the glue is at least thoughtfully applied.