Sound design pass

Prompt

Do a sound design pass on [CompId]. The picture is locked; do not change visual timing unless a sound cue proves a cut is badly placed, and then ask first.

1. Cue sheet first. Go through the storyboard and the actual composition and list every visual event that deserves sound: scene changes, the main reveal, impacts, clicks, counters landing, the end card. For each: frame, sound type (whoosh, impact, pop, click, tick, riser, swell, silence), level, and why. Fewer is better: transitions, interactions and the payoff. Silence is a choice; mark where it helps. Show me the cue sheet and wait.

2. Music bed: [music file or "propose options"], mood [mood], tempo [bpm]. Arrangement follows the story: sparse under the problem or question, lift into the payoff, an impact on the main reveal, a clean ending on the end card with a fade over the last 1 to 1.5 s or a natural tail. Prefer landing scene changes on bar lines; tell me where the music and the cuts disagree.

3. Sound effects: place each by its peak, not its start. If the whoosh peaks 0.4 s into the file, it starts 0.4 s before the visual hit. Click sounds lead the visual UI reaction by 80 to 150 ms. Keep SFX consistent in character across the film.

4. Voice-over: [VO file or none]. Duck the music under the voice (about -10 to -12 dB while speaking, 150 to 400 ms release so it does not pump). The voice must always be intelligible.

5. Mix and measure: target [-14 LUFS integrated, true peak at or below -1 dBTP] for web and social ([-16 LUFS] for presentations). Measure, do not guess: export the mix and run ffmpeg -i [mix.wav] -af ebur128=peak=true -f null - , and report integrated loudness, LRA and true peak. Normalise with loudnorm=I=-14:TP=-1:LRA=11 if needed.

6. Report every audio file used, its source and licence. Flag anything unlicensed or unclear.

7. Derive, do not hand-place: compute each cue frame from the same constants and spring functions the picture uses (the contact frame of a spring, the 50 percent frame of a pop-in, the velocity peak of a whip). Retime the animation and the sound retimes with it. Verify with the sync waveform check afterwards.
8. Restraint and silence: one soft sound per text fragment, near-silence while it is read, one big hit for the one reveal. Spotlight a key claim with a short dropout in the music (a few frames to half a second of hole, then the beat returns on the next visual event) instead of another effect. A calm film may be effects-only.
9. Beat-hit cap (music-led films): beat sync does not mean pulsing the whole picture. At most 3 full-frame or camera-layer impacts per film, at least 16 beats apart; every other hit lands on the hero element only.
10. Grid acceptance (track you did not write): fit a straight line to the detected beats, report the residual (accept up to about 15 ms), test half-time and double-time against where the kicks fall, keep float seconds until the last step into frames, and after render measure cut error (target 3 frames or less). Details in 75.

Use deterministic timing: all audio placement in frames, derived from the same constants as the visuals so cues move with the edit.