You give it a video file, an audio track, and a description of what you want, and it uses Gemini's multimodal API to generate a beat-synced edit plan that maps video clips to the music's rhythm. The workflow is remarkably straightforward: a bash script uploads both files to Gemini, gets back a JSON EditPlan with timestamps and clip selections, then pipes that through validation and FFmpeg filter generation scripts. The whole thing spits out a proper filter_complex command for rendering. It keeps uploaded files cached for 48 hours by default so you can iterate on prompts without re-uploading, which is smart. If you've ever manually scrubbed through footage trying to find moments that hit on beat drops, this automates exactly that part.
npx -y skills add ecliptic-ai/skills --skill beat-sync-video-editing --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
juliusbrussee/caveman
mattpocock/skills
shadcn/improve
obra/superpowers
forrestchang/andrej-karpathy-skills
vercel-labs/skills