Sonvane

Choosing music for a video involves more than finding a good track. Creators must decide when it should enter, whether it leaves room for the original sound, and which version supports the scene. **Sonvane** connects those decisions in one AI-assisted workflow. Meta Muse Spark reviews sampled video frames, available original audio, and the creator's request, then proposes an editable plan for music and sound effects. The creator reviews and approves the plan before ACE-Step generates music variations. TuneJury evaluates the generated music. Sonvane combines its preference scores with text-to-music matching and explicitly saved listening preferences to recommend a soundtrack. Creators can compare versions at the same playback position, choose a different track, revise a passage, adjust the mix, and export their video. The contribution is a working connection between scene understanding, human approval, generation, evaluation, and personal choice. Existing model weights remain unchanged; saved comparisons update a separate personal ranking model. Try Sonvane: https://yonghyunk1m-sonvane.hf.space/demo/