| name | suggest-sfx |
| description | The SFX pass. Analyze a short's beats + narration and propose tasteful sound effects synced to them, drawing from (and growing) a shared, reusable SFX library, then render an SFX-mixed audition preview. Use when the user wants to "add SFX / sound effects", "suggest sfx", "score the transitions", "sound-design this beat", generate/source sound effects, build or extend the sfx library/catalog, author or audit a sfx-plan, or mix SFX over a rendered short in this repo. Covers reading beats + word times + brand §7, the library-first flow, generating misses with the ElevenLabs Sound Effects API, the per-video sfx-plan.json, the hard user-audit gate, and mixing with tools/mix_sfx.py (light voice ducking). |
suggest-sfx — the SFX pass
Take a rendered, voiced short and add tasteful sound effects, synced to the visual beats and
the narration, drawn from a shared, growing SFX library. Work collaboratively and
beat-by-beat; the user audits the plan and gives notes.
The shape: a declarative plan (sfx-plan.json) is the source of truth → a tool consumes it
(tools/mix_sfx.py) → a library grows (media/library/sfx/) → a hard USER-AUDIT gate
before anything is mixed.
Sound taste is a brand contract — read brand.md §7 every time. The house style is calm/premium
(Linear / Anthropic / Vercel), felt-not-heard, always under the voice. NOT MrBeast-loud.
Inputs (read these first, every session)
brand.md §7 (Sound design) — the SFX taste contract: subtlety, density, palette, levels,
sync, signature motifs, source policy. Non-negotiable.
shorts/short-N-<niche>/beats.json — the beats and REAL per-word voice times (written back
by gen_voice.py). Sync cues to these words (a pop on "boom", a chime on "free").
- The shot source
remotion/src/shots/short-N/*.tsx — the INTERNAL animation frames are where
the real visual beats are (piece lands, digit pops, badge stamps). Read the shot, convert its
local frame to global time: at_s = (local_frame + sequence_from) / fps. Do not guess — the
cue must land on the exact frame the thing happens.
media/library/sfx/catalog.json + palette.json — the library you draw from and grow.
The library (media/library/sfx/) — the durable asset
palette.json — generation recipes: generic, reusable sound ids + prompts + duration + tags.
The source of truth for what the library SHOULD contain.
catalog.json — the manifest of what EXISTS: {id, file, category, tags, duration_s, peak_dbfs, loudness_lufs, source, model, license, prompt, used_in} per clip. Written by gen_sfx.py.
clips/*.mp3 — loudness-normalized clips (~−20 LUFS, −1.5 dBFS ceiling) so a plan's per-cue
gain_db is perceptually meaningful.
- Library-first, always. Reuse an existing clip before generating. Name/tag clips GENERICALLY
(
ui-toggle-on, whoosh-soft, pop-reveal) so future videos reuse them — the library is the
durable asset, each video is one draw from it. Only genuinely-missing sounds get added to
palette.json and generated.
Workflow
- Read brand §7 + beats.json + the shot source + catalog. Pull exact cue times: for each
candidate beat, find the visual moment's local frame in the shot and/or the narration word,
and convert to global seconds.
- Decide taste with the user up front if unsettled (present-ness, density, which kinds,
source) — then apply brand §7. Don't re-ask settled decisions.
- Propose the cue list — function-first. For each beat ask what it needs (brand §7): motion
(whoosh) · tension (riser) · emphasis (impact/pop) · snap (click). Score key
transitions, reveals, and tasteful click-sequences; a 3–4-shot click-sequence counts as ONE
gesture. Mark genuinely deniable texture as
"optional": true. Layer the 2–3 biggest moments
(build-and-drop): riser→impact on a scripted reveal, whoosh→pop so a cut stands out — in the
plan a layer is two events at the same/adjacent at_s that sum. Sync each cue to the VISUAL
beat and often the exact word.
- Library-first sourcing. For each distinct sound the plan needs, reuse a catalog clip if one
fits. For misses, add a generic recipe to
palette.json and generate: python tools/gen_sfx.py
(ElevenLabs Sound Effects API → normalized clip → catalog). Fall back to curated royalty-free
only where generation is weak; record source + license either way.
- Author
shorts/short-N-<niche>/sfx-plan.json — one event per cue, at_s in global seconds.
(Schema below.)
- USER-AUDIT GATE (hard). Present the resolved cue sheet
(
python tools/mix_sfx.py <plan> --print) and get the user's approval/notes BEFORE mixing.
- Mix the audition preview once approved:
python tools/mix_sfx.py shorts/short-N-<niche>/sfx-plan.json
→ the plan's render.out. Light sidechain duck under the voice + a safety limiter. Iterate on
the user's notes (gains, timing, add/cut cues) — re-print, re-mix.
- Verify audibility with numbers, not hope. After mixing, RMS-diff the mixed audio vs the
voice-only preview at each cue window: a story-critical cue should add ≥ +4 dB, texture
+1–3 dB. Inaudible cues hide in cue sheets — don't ship a cue you haven't confirmed lands.
- Transient-clip gain gotcha. Short percussive clips (knock/stamp/keys/snap) hit the
−1.5 dBFS peak ceiling BEFORE reaching the −20 LUFS loudness target, so they catalog ~3–5 dB
quieter than sustained clips. Their plan must be than the brand
table implies. The audibility check above is what surfaces this.
sfx-plan.json (the source of truth)
{
"master": "remotion/out/ShortNName-voiced.mp4", "master_fps": 30,
"catalog": "media/library/sfx/catalog.json",
"render": {
"preview": "remotion/out/ShortNName-voiced.mp4",
"out": "shorts/short-N-<niche>/output/short-N-sfx.mp4",
"end_s": 42, "duck": true
},
"events": [
{ "at_s": 4.9, "sfx_id": "chess-piece-thock", "gain_db": -8, "shot": "MainScene",
"cue":
at_s — global-timeline seconds (same clock as beats.json).
gain_db — dB relative to the clip's normalized level (lower = quieter). See the SHORTS
calibration above for starting values.
optional: true — deniable texture; mix_sfx.py --no-optional drops it. Use it liberally so
the audit is about the core set.
cue / note — human-readable anchor (the exact frame/word) so the audit is legible.
Tooling quick reference
- Grow the library:
python tools/gen_sfx.py [--dry-run] [--only id1,id2] [--force] [--renorm].
--renorm re-balances existing clips to the loudness target (no API/billing). Needs
ELEVENLABS_API_KEY in .env.
- Print the cue sheet (the audit artifact):
python tools/mix_sfx.py <plan> --print [--no-optional].
- Mix:
python tools/mix_sfx.py <plan> [--no-optional] [--no-duck] [--end S] [--out path].
Principles (the house style — apply them)
- Under the voice, always. SFX are seasoning; the voice is the show. Duck them, keep them
quiet, never let one peak above the narration. Silence is part of the mix.
- Key transitions & reveals only — plus the diegetic action sounds that ARE a synthetic short's
content (piece thocks, digit pops). Everything past that is
optional.
- Sync to the visual beat AND the word. A click exactly on the toggle flip; a pop exactly on
the reveal. Off-by-100ms reads as sloppy — use real frame/word times.
- Reusable-first. Generic names + tags so the library compounds across videos. Reuse before
you generate. The library is the deliverable that outlives this video.
- Function over vibe. Pick the sound by its job — motion/tension/emphasis/snap (brand §7) —
not by browsing the catalog for something that "feels right." The same few foundational sounds
do the heavy lifting; more is not better. Keep signature motifs consistent.
- Layer the big moments, keep the rest single. Build-and-drop (riser→impact, whoosh→pop) is
where SFX earn their keep — but only on the 2–3 hero beats.
- Take structure, not drama. Calm/premium: no cymbal-urgency risers, no trailer slams, no
whoosh on every move.
- Audit before mixing. The user reads and approves
sfx-plan.json first. Non-negotiable.
Done = the library has the needed clips (catalogued with source+license), sfx-plan.json is
authored and audited by the user, the SFX-mixed preview is rendered and spot-checked by ear,
and any new clips + used_in are saved back to the library.