# Apex VIP Pro Tips (synced)

- VFR: -framerate 30 on still-image inputs
- A/V: apad+shortest; -ar 48000; drift < 0.5s
- Dynamic captions word-by-word = highest short-form impact
- Hard cut ~90% of edits; transitions only when story needs them
- Hook cut every 1–2s first 3s; listicle: new visual every fact
- visual_judge 5 axes → FAILED/PROPER/EXCELLENT; block done on FAILED
- Brain orchestrator 3-pass: intent brief → sub-agents → editorial review
- Theme seeds: reuse EXCELLENT recipes before inventing style
- VLM critique free-first; offline → PROPER+skipped never crash
- EBU R128 loudnorm I=-14:LRA=11:TP=-1 after mux
- Never invent metadata; probe or mark INCONCLUSIVE
- Max 3 re-render attempts; checkpoint after every stage
- H.264 social default; face blur when privacy requires
- Load expert-editor-logic + brain_orchestrator + expert_tips + effects and edits + visual_judge + vlm_critique

## Skill functions
- **boot_check**: Load skill + permanent JSON + pro-tips BEFORE any edit/render
- **perceive**: Hybrid perception: native vision for meaning; scripts only for numeric metrics
- **metadata_detect**: ffprobe format+streams; image EXIF/XMP/IPTC; never invent duration/size
- **vfr_guard**: Always -framerate 30 on image inputs
- **av_sync**: apad + -shortest; final audio -ar 48000; drift < 0.5s
- **visual_judge**: Run visual_judge.py → FAILED|PROPER|EXCELLENT; block done on FAILED
- **brain_orchestrate**: 3-pass brain orchestrator + intent brief (brain_orchestrator.md)
- **theme_seeds**: Check theme_seeds before style; capture EXCELLENT recipes
- **vlm_critique**: vlm_critique.py after pixel pass; free-first cascade
- **loudnorm**: EBU R128 loudnorm=I=-14:LRA=11:TP=-1 after mux
- **state_checkpoints**: StateEngine after every stage; max 3 re-render attempts
- **qa_gate**: 5-tier QA + final_gate; never promote INCONCLUSIVE
- **effects_kit**: effects and edits.md: hard cut ~90%; tone-matched effects; cut rhythm by type
- **faceless_rules**: NO EYES/NO FACE FEATURES when faceless brand content
- **modes**: --mode ad (5) or narrative (7); --ai-clips before Ken Burns