Vibe Film
Turn a "hardcore concept × the thing you think about at night" into a no-voiceover knowledge mood short film: viewers read bilingual subtitles, watch the scene act out the sentence, and end with a line that hits — "Dedicated to everyone who __".
Division of labor: Human’s job = topic gate + shot-by-shot table + direction approval (BGM demo / style frames / ability demo samples) + sign-off. Model’s job = reverse benchmarking + code + render + self-check and rework. Approval items follow rework cost: expensive and hard-to-revert items must be gated (storyboard / style frames / demo / BGM / publish). Programmatically checkable items don’t require human review (technical checks / intra-segment self-checks). Quality comes from storyboard specs and approval gates, not from model improvisation.
Where good films come from: three layers of content density, word-by-word subtitle rhythm, a manually calculable number, audiovisual alignment — but exactly how it moves and how much is decided by the storyboard and emotional arc. The density baseline (storyboard.md) is the floor, locked in. Technique density and variation are the ceiling, open. Everything beyond the baseline is the model’s starting point, not its limit.
Applicability
| What the user wants | Use this? |
|---|---|
| No-voiceover knowledge mood film (concept × emotion, bilingual subtitles, music only) | ✅ |
| "2AM–vibe" short film on philosophy/science concepts | ✅ |
| Voiceover explainer, line-by-line teaching, tutorial screencast | ❌ Voiceover pipeline (TTS)‑driven |
| Editing/mixing existing footage | ❌ |
| Predominantly live‑action footage (>50% real/photo‑real feel) | ❌ Hybrid route (generated assets + code animation), needs discussion |
Duration: 90–135s per project (1–4 segments, each ≤45s). 2:40+ extends via 4‑segment scheme. Vertical format not supported (composition natively horizontal).
Model Strengths and Default Pitfalls
- Strengths: Zero engineering failures under spec‑driven long prompts (tested: four‑segment sub‑agent had zero structural rework); frame‑by‑frame deterministic rendering; can work for extended periods on “render static frame → review → modify”; reliable visual detail execution within the nine‑column constraint.
- Default pitfalls: When given only a one‑sentence shot intent, the scene falls into a “moving PPT” with a single subject and changing subtitles; without reference frames, aesthetic quality lacks an anchor; free‑form generat...