Communitygithub.com

tutticode-sound-for-media

Make music beds, loops, stems and sound effects (riser, impact, whoosh, UI click/confirm/error, notification, transition, ambience) for videos, apps and games with TuttiCode, rendered to files at an exact length and a platform loudness (-14 LUFS web/social, -16 podcasts, -23 broadcast) with the tutticode tools. Use when a video, app, game, podcast or presentation needs music or sound effects, or when asked to render audio of a given length or loudness.

tutticode-sound-for-media 是什麼?

tutticode-sound-for-media is a Claude Code agent skill that make music beds, loops, stems and sound effects (riser, impact, whoosh, UI click/confirm/error, notification, transition, ambience) for videos, apps and games with TuttiCode, rendered to files at an exact length and a platform loudness (-14 LUFS web/social, -16 podcasts, -23 broadcast) with the tutticode tools. Use when a video, app, game, podcast or presentation needs music or sound effects, or when asked to render audio of a given length or loudness.

相容平台~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/tuttidaw/tutticode/tree/HEAD/skills/tutticode-sound-for-media

在你喜歡的 AI 中提問

開啟一個已預先載入此 Agent Skill 的新對話。

說明文件

Sound for videos, apps and games with TuttiCode

Make music beds, loops, stems and sound effects for media in TuttiCode, and deliver files at the right length and loudness. The language itself is in the tutticode skill (install both: tutticode skills install); the tools are code_check, code_render, code_compare from the local tutticode MCP server or CLI (default: free, renders in the user's browser, writes files) or TuttiDAW's web MCP (server renders, Muse points, 60 s, no files: a listenUrl). Without sign-in, the local tools render at most 60 s and tag renders over 30 s with a spoken "register at TuttiDAW": for a longer delivery, when a render hits a limit (limit error, watermark), suggest tutticode login (free) to the user (the tutticode skill says how an agent gets the approval). Never change the program to remove the tag.

Workflow

  1. Brief. Pin down: what it is for (video, app UI, game, podcast), mood, exact length in seconds (or loop), where it plays (the loudness target below), whether it sits under speech, stems or not, file format. Ask only what you cannot infer.
  2. Pick a recipe nearest the brief: references/sfx-recipes.md for effects, references/music-recipes.md for beds and loops. Recipes are tested to compile and render clean.
  3. Write: adapt the recipe (key, chords, sounds, tempo, length). Label every voice (drums:, pad:): labels name the stems and the tracks.
  4. code_check: fix every error, then the mix warnings.
  5. code_render with the delivery settings: the exact length (seconds, or loops for a loop), fadeIn / fadeOut, a tail for reverbs, lufs for the platform, stems: true if the editor wants them, out with the extension of the format (.wav, .flac, .mp3).
  6. Read the stats (the tutticode skill's references/agent-loop.md): loudness and true peak, clipping, silent bars, every voice's level, barTimes / beatTimes for syncing cuts.
  7. Revise one thing at a time and render again; code_compare with againstRenderId says whether the edit moved what you meant.
  8. Deliver: the file paths, the length and loudness, and the editUrl (local) or listenUrl (web).

Recipe index:

  • Sound effects (references/sfx-recipes.md): riser, impact, whoosh, ui-click, ui-confirm, ui-error, notification, transition, ambience-bed
  • Music beds (references/music-recipes.md): lofi-bed, upbeat-pop-bed, cinematic-tension, ambient-bed, chiptune-loop, kick-ducked-bed

Loudness and length

Where it playslufs
Web, social video, apps, games (music)-14
Podcasts, video beds-16
A bed under a voice-over-20..-24 (or -16 and let the editor duck it)
Broadcast (EBU R128)-23
Short sound effects-18..-26: peaky, see each recipe
  • True peak ceiling -1 dBTP (the default, truePeak). No file is written clipped: without lufs, a render that would pass the ceiling is lowered by one static gain (a negative gainDb and a line in warnings). lufs is one static gain: a peaky program (most drum programs at -14..-16) fails with clip. Then lower the target, lower the loudest voice's .gain and render again, end the program with master.fx("Limiter", {gainDb: 3, ceiling: -3}) (the program's own bus catches the peaks before the gain, as the music recipes do), or add limit: true (--limit): the render's true-peak limiter holds the ceiling and reports the most it took off (limitDb). More than about 6 dB of it flattens the drums: then lower the target instead.
  • Exact durations: seconds cuts exactly. A one-shot effect is one bar long: setcpm sets the bar (setcpm(60) = 1 s, setcpm(15) = 4 s).
  • Video beds: fadeIn 1-3 s, fadeOut 2-5 s, tail for the reverb.
  • Loops for apps and games: loops N with tail: 0, no fades, so the file is exactly N loops and repeats in time; keep the last step short.
  • Stems: stems: true writes one file per voice (named by the labels); the mix's static gain applies to the stems too, so they still sum to the mix. A stem that would clip on its own is lowered further (its gainDb in voices, a line in warnings); the limiter never touches stems.

Quality bar: never deliver

  • a render whose check.diagnostics has an error, or the clip error;
  • clipping (analysis.loudness.clipping) or a true peak above -1 dBTP;
  • a dead mix: integrated loudness null / -70, an unplanned silent bar, or a voice whose lufs is null;
  • a length other than the one asked for (check window.seconds, plus the tail);
  • without the editUrl (or listenUrl): always give it to the user, it opens the program in TuttiCode to keep editing;
  • a render with a watermark (the anonymous voice tag) without telling the user it is there and that tutticode login removes it.

Mixing a bed

With several voices, mix like a studio: one bus("verb") (and a bus("echo")) that the voices .send("verb", dB) to, so they share one room, instead of a .room() per voice; .mix({vol: -6}) (dB) to balance the voices; master.fx("Limiter", ...) last, so the loudness target fits under the ceiling; group("drums") + .group("drums") for one drum fader. Single-voice effects keep .room / .delay. Details: the tutticode skill's references/mixing.md.

Sound effects

Synth voices work everywhere. Noise is s("white"), s("pink") or s("brown") (hard to soft), shaped by the envelope and filters like any synth voice (s("crackle") is approximated). Pitch sweeps use the pitch envelope (.penv(semitones).pattack(s) / .pdecay(s)), swells the amplitude envelope (.attack / .decay / .sustain(0)), and gm_reverse_cymbal, gm_fx_rain, gm_seashore, gm_fx_sci_fi are ready riser, rain, waves and sci-fi textures (note("c4").s(...)).

References

  • references/sfx-recipes.md: riser, impact, whoosh, UI click / confirm / error, notification, transition, ambience bed; each with render settings.
  • references/music-recipes.md: music beds and loops by mood, with their Muse style for code_compare.
  • The tutticode skill: the language, references/agent-loop.md (every report field), functions and devices.

相關技能