Riffmaw
Riffmaw turns remembered moments, recordings, words, and musical intent into attributable music. AI generation, instrumental or vocal performance, editing, and signal processing can all contribute; the Magus remains the producer and no model or DAW owns the craft.
| Field | Reference contract |
|---|---|
| Identity | riffmaw.music revision 1 |
| Principal Pattern | riffmaw.forge_music_bundle@1 |
| Begins with | a music brief plus any explicitly admitted references, lyrics, takes, MIDI, scores, samples, or live inputs |
| Can return | PerformanceTake@1, MusicalPerformanceSession@1, MusicAssetBundle@1, MusicCandidateSet@1, MusicFindingSet@1, optional MusicCueMap@1, or an exact refusal |
| Stops before | ordinary dialogue, spoken dialogue replacement or spoken-media localization, picture-bound sound design, visual generation, final audiovisual assembly, performer impersonation, or publication |
The earlier Designed-only riffmaw.audio revision 1, its Sonic* contract family, and the former
riffmaw.forge_track@1, riffmaw.capture_take@1, riffmaw.open_jam@1,
riffmaw.index_sources@1, and riffmaw.mark_moment@1 meanings are retired, not reinterpreted.
riffmaw.music revision 1 is a new, narrower identity and uses those five Pattern names only as
@2 revisions. No Portfolio registry, Run, or stored application record uses the retired design, so
there is no executable migration; historical references retain their old meaning.
Prompt language, lyric language, requested sung language, and interface locale remain separate. The exact Riffmaw model profile owns proved musical and language behavior; Soulstone may bind it to a local ComfyUI or another admitted runtime, while Portal may bind a separately evidenced hosted API. Similar product names never make those implementations equivalent. Music owns the request boundary and Studio keeps the candidate routes.
Ways of making
Riffmaw can complete the same application purpose through different production modes:
| Mode | Representative Pattern | Human and machine relation | Typed finish |
|---|---|---|---|
| Forge | riffmaw.forge_track@2 |
A brief, references, lyrics, or an empty session drive offline composition and generation. | MusicCandidateSet@1 or MusicFindingSet@1. |
| Perform | riffmaw.capture_take@2 |
Explicitly armed audio or MIDI preserves what one or more people play, sing, rap, perform as musical spoken word, or program. | One attributable PerformanceTake@1 plus exact observations or findings. |
| Jam | riffmaw.open_jam@2 |
Performers and admitted machine players share mapped clocks and answer one another live or ahead of a declared musical boundary. | Recoverable MusicalPerformanceSession@1 or MusicFindingSet@1. |
| Produce | riffmaw.forge_music_bundle@1 |
Riffmaw iterates arrangement, mix, mastering, deterministic probes, and attributable listening review. | MusicAssetBundle@1, MusicCandidateSet@1, MusicFindingSet@1, or refusal. |
The Patterns connect through typed artifacts and new forward Invocations; one Pattern never nests or silently resumes another. A later Scroll may publish exact capture, generation, render, and review Spell placements inside one casting, but only with its own revision, Resolution Lock, and authority ceilings. Opening a file does not arm an input; arming a guitar does not authorize synthesis or model training; accepting a machine accompaniment does not approve the mix; exporting a master does not authorize publication.
The work
- Sources finds exact moments without turning an ambient media library into authority.
- Sessions preserves armed inputs, human recordings, takes, MIDI, live jams, and performance conditions.
- Music covers composition, AI generation, arrangement, editing, mixing, and mastering.
- Studio defines the replaceable production-tool roles and the current candidate stack.
- Voice keeps musical vocals here while routing ordinary dialogue, spoken dialogue replacement, captions, and spoken timed-language editions to Language Edition.
- Sound distinguishes musical production effects from Broadcast's picture-bound sound and the unresolved reusable-sound office.
- Music hardware boundary admits exact instrument and musical-vocal capture routes while the shared Animator page keeps generic Linux, speech, mobile, and body mechanics.
- Sync exposes musical timing without directing the image.
- Package seals lineage, effects, recovery, and the consumer handoff.
Voidlight owns visual assets. Broadcast owns picture-bound sound, editorial assembly, the final timeline, and publication effects. Language Edition owns ordinary dialogue, spoken dialogue replacement, captions, and spoken timed-language editions. Reusable standalone sound remains an unresolved future standalone-sound Composition boundary rather than hidden Riffmaw scope.
Related: Audio · Workflow · Composition portfolio