Current formats
Dates show the last revision of the format description, not the platform's release date.
| Engine | Model / version | Fields briefed | Revised |
|---|---|---|---|
| Eleven Music | elevenlabs_music | Prompt · composition plan | 2026-10-22 |
| Suno | v6 | Style · lyrics · exclude styles · two sliders | 2026-11-10 |
| Udio | udio | Describe · lyrics · seed · clarity | 2026-11-10 |
| Mureka | V9.5 | Prompt · styles · lyrics | 2026-11-10 |
| MiniMax Music | Music 3 | Prompt · section-tagged lyrics · flags | 2026-11-10 |
| YuE2 | YuE2-3B | Lyrics · style prompt · genre tag | 2026-11-10 |
| Soundverse | v9 | Describe (≤500) · tags · lyrics | 2026-11-10 |
| Treblo | Melodia v3 | Prompt · tags · lyrics · prompt strength | 2026-11-10 |
| Stable Audio | 2.5 | Prompt · steps · CFG · seed | 2026-10-22 |
| Soundraw | — | Genre · mood · tempo · length | 2026-10-22 |
| Mubert | — | Prompt · tags · track/stream mode | 2026-10-22 |
| Beatoven | — | Mood session · timeline edits | 2026-10-22 |
| AIVA | — | Style preset · key and tempo · editor passes | 2026-10-22 |
| Loudly | MANTA 1 / VEGA | Style (≤1000) · lyrics (≤3000) · vocal gender · instrumental flag | 2026-11-10 |
| MusicGen | AudioCraft large | Description · model size · sampling parameters | 2026-11-10 |
Formats are reviewed when a platform ships a model or field change, and whenever one of our own briefs is corrected. A correction is never made quietly: it lands here with a date.
History
- 2026-11-10 — Loudly corrected. MANTA 1 is confirmed as its vocal model, so the entry and guide now cover songs with lyrics as well as instrumental beds.
- 2026-11-10 — Engines 12–15 added: Soundverse, Treblo, Loudly and MusicGen, each with a guide page and a JSON payload branch.
- 2026-11-10 — Engines 7–11 added: Suno v6, Udio, Mureka V9.5, MiniMax Music 3 and YuE2. Suno's exclusion is lifted at the operator's request; output remains text only.
- 2026-10-22 — Initial six formats: Eleven Music, Stable Audio, Soundraw, Mubert, Beatoven and AIVA.
Developer payload schema
The API JSON view returns one object per engine. The schema string is stable within a major version, and fields are only ever added — so a script that pins the schema keeps working.
brief_id is derived from the measurements and the engine, so the same analysis of the same file returns the same id — useful for de-duplicating a batch without storing the audio.
source carries the measured facts, so a pipeline can re-derive a brief without re-running the browser analysis.
model_engine is retained from schema 1.0's first release and will not move.
Where this is enforced
The same commitments appear in the Trust & Content Policy: text only, no audio output, no sampling, no voice models, no artist or song matching. A format change cannot cross those lines, whatever a platform adds.
See The Current Formats
Open the Studio, drop a reference, and pick an engine — the brief you get is the one described in this table.
OPEN THE STUDIO →