Skip to content

Autopilot

Autopilot is Onset Engine’s one-click pipeline. Point it at a folder of footage and a music track, and it delivers a fully rendered, beat-synced video — no manual timeline work required.

Input: footage_folder/ + track.mp3
Output: output_clips/output_STANDARD.mp4 (30fps, full VFX)

Autopilot runs through 5 stages, each emitting [STAGE] events that drive the GUI’s progress stepper:

StageNameWhat Happens
1INGESTINGScans footage folder, runs CLIP analysis on every clip (or detects already-ingested clips)
2ANALYZINGProcesses the music track — beats, energy curve, drop detection via librosa
3SELECTINGMatches clips to musical energy using driver scoring (contrastive similarity)
4GENERATINGCreates a draft EDL (Edit Decision List) — clip placement + transitions
5RENDERINGFull-quality render at 30fps with VFX, transitions, and final audio mix

Instead of writing a JSON driver, you can describe your creative vision in plain text:

  • “During quiet parts, focus on:” → maps to the 1_LOW tier
  • “On the heavy drops, focus on:” → maps to 3_HIGH and 4_MAX tiers

Autopilot auto-generates a 4-tier driver from these descriptions via PromptParser._generate_driver(). The generated driver is saved to drivers/_autopilot_<hash>.json for cache reuse.

Priority chain: Explicit driver JSON → text descriptions → no driver (motion-score fallback).

See Text Descriptions for the full guide.

During rendering, the GUI displays a live timeline showing clips being placed in real-time:

  • [TIMELINE_INIT] events set up the canvas with duration and beat markers
  • [CLIP_PLACED] events add colored blocks as each clip is committed to the timeline
  • Clip colors correspond to energy tiers (purple → blue → teal → pink)
  1. Open the Autopilot tab in the right panel
  2. Select a source folder and music track
  3. Optionally enter text descriptions for clip direction
  4. Click 🚀 GENERATE & RENDER
  5. Watch the Glass Engine timeline fill in real-time
  6. Final video appears in your output directory