Scenario Sonilo Audio

Overview

Sonilo, an audio family on Scenario, splits along two axes: what the sound is (sound effects or instrumental music) and what comes back (a standalone track, or the source video with the generated track muxed in, visuals untouched). Picking the member whose return matches the delivery matters more than any prompt. Discover with search and treat model_schema_get as the contract.

Connection and the core loop: see the scenario skill in this repo; model-agnostic audio work: the scenario-audio skill. If a sibling skill named here is missing from your available skills, ask the user to install it (npx skills add scenario-labs/skills --skill <name>); unattended, proceed from tool schemas and flag the gap.

Quick reference

Member names state the mode; input names come from the live schemas:

Member	Key inputs	Returns
Text to SFX	prompt, duration, audioFormat	one sound effect clip
Text to Music	prompt, duration, numSamples	instrumental track
Video to SFX	video, optional prompt or segments	audio track matching the video length
Video to Music	video, optional prompt, numSamples, startOffset, duration	music track
Video to Video SFX	as Video to SFX	the video with SFX mixed in, plus the track
Video to Video (music)	as Video to Music, plus keepSpeechVocal	the video with music mixed in

At authoring time: text SFX ran 1 to 180 seconds (default 8), text music 1 to 600 (default 90), video inputs took up to 360 seconds, segments up to 50, numSamples 1 to 3, and startOffset moved in steps of 10 with startOffset plus duration capped at the video length. audioFormat (aac default, mp3, wav, flac; wav or flac for editing pipelines) exists on the SFX members only, and on the muxed one it formats the separate track: the video's own audio stays AAC. Two music mux members were live with the same schema; prefer the newest hit. Every duration knob carries cost, so dry_run before a batch.

The video is the clock

On video-conditioned members the footage decides when sound happens and the prompt only steers what it sounds like; the clip's own audio never steers generation, visuals alone are read. An empty prompt is valid and often best: the model captions the clip and covers each scene itself. When one description cannot fit the whole clip, segments gives per-range prompts, contiguous by contract: the first start is 0, each end equals the next start, and the last stays within the video. Text to SFX has no timing control at all, so describe one sound event per run, physical words over moods (hollow, muffled, punchy), source then material then space then intensity, with the duration intent in the wording as well as the parameter. No member takes a seed, so archive the takes you like; on the music members numSamples buys up to three takes in one run instead of re-rolls.

What survives a music pass

The music mux members replace the whole original track by default. keepSpeechVocal: true isolates human voice (dialogue, narration, singing, crowds) and ducks the music under it; everything non-voice (engines, footsteps, ambience, prior foley) is replaced regardless. So never score a clip after muxing foley into it: when a video needs both, take standalone tracks from Video to SFX and Video to Music and mix in post. Music comes back instrumental.

Worked example: foley for a silent gameplay clip

1. search with target="models", query="sonilo", public=true. Members list txt2audio, video2audio, or video2video capabilities; e.g. model_sonilo-v1-1-video-to-video-sound-effects (a live hit at authoring time: re-discover each session).
2. model_schema_get with that id: fields and caps before anything else.
3. upload_asset the clip (see the scenario skill) to get its asset id.
4. model_run with that model_id, dry_run=true, and parameters={"video": "asset_x", "segments": [{"start": 0, "end": 4, "prompt": "footsteps on wet metal, close and sharp"}, {"start": 4, "end": 12, "prompt": "plasma rifle shots, hollow hangar reverb"}]}: cost scales with clip length.
5. Repeat model_run with wait=false, then jobs_wait with the returned job id, re-called with pending_job_ids on timeout, never a second model_run.
6. asset_display the muxed video, then asset_download it and the separate track.

Common mistakes

• Muxing when the edit needs a bare track, or the reverse: Video to SFX returns audio only; Video to Video SFX returns the finished clip. Pick by delivery.
• Scoring a clip that already carries foley with a music mux member: non-voice sound is replaced and the foley is gone.
• Leaving keepSpeechVocal off (the default) on a talking-head clip: the narration vanishes with the rest of the track.
• Gapped or overlapping segments: the contract wants contiguous ranges from 0.
• Packing sequenced events into one Text to SFX prompt: it has no timing control; one event per run.
• Asking any member for speech or vocals: no member produces dialogue, narration, or singing; SFX and instrumental music are the whole surface.