37 · Music · Creative and media · live
Rights-cleared music generation
Eval results
Scored on a held-out or test splitRun 25 Sep 2026Eval write-up (decosa-api, access required)
- Prompt guard precision (refuse)100% (48/48)test splitn = 79Dev: 96.3% (26/27)
- Prompt guard recall (refuse)98.0% (48/49)test splitn = 79Overt 26/26, sneaky 22/23; code rules alone refuse 55% (27 of 49).
- Safe prompts refused0 of 30test splitn = 30Dev: 1 of 15 (a public-domain composer refused).
- Planted near-copies flagged at the flag threshold80.8%test splitn = 24090.4% at the review threshold. About one planted copy in five slips under the flag.
- False flags on unrelated tracks by new artists2.3% (2)test splitn = 88Held-out pool tracks: 1.9% (1 of 53); fresh renders 0 of 9.
Dataset
Prompt guard: 120 hand-written synthetic prompts (45 safe, 40 overt, 35 sneaky), 41 dev / 79 test, written before the guard ran. Similarity: 240 commercially licensed Free Music Archive tracks as the catalogue; 64 planted near-copies x 6 edits (24 dev / 40 test sources), plus unrelated and held-out negatives; thresholds set on dev only, applied once to test.
Caveats
- The guard prompts were written by hand for this eval by the builder; synthetic.
- The similarity check is a triage signal for near-copies of the references it holds, not a clearance: 240 Creative Commons tracks are not the recordings anyone is likely to copy.
- Lyrics are not compared.
- Music quality is not measured.
- Guard latency (median 30.3 s) was measured on a saturated shared gateway.
Nightly smoke check
Loading the nightly status…
- Result
- pass
- Run
- 25 Sep 2026
- Latency, this run
- n/a
- p50 over passed runs
- 43 s
- Receipts
- 1
- Model calls
- n/a
- Tokens
- n/a
- Cost per run
- <$0.001
Self-host verification
Verified on 25 Sep 2026: Fresh clone of the branch into a clean directory, docker build of the api image plus the documented FFmpeg and c2pa-python layer and of services/music_embed, compose with named volumes, pointed at the already-running local vLLM (direct route) and ComfyUI; then torn down.
The guard refused and stripped as documented, a 30 s MiniMax-Music3 render finished in 21.4 s with a similarity verdict and a certificate that verified (and failed when edited), a planted near-copy scored 11.6 against the catalogue, and no prompt text reached the logs. Found on the way: the api image has no c2pa-python, so the first render had no C2PA credential; with the layer and the dev certificate, stamping inside the container produced a credential that validates.
Rehearsal bundle: music-gen-cleared.zip (506 KB, 12 checks). Mock inputs plus the expected results, so you can prove your own setup works before any real data touches it.
Known limits
- The similarity check knows only 240 Creative Commons tracks and the files you compare; it flagged 81% of planted near-copies at 2% false alarms, and a clear result says nothing about commercial catalogues.
- The guard reads text only: 98% recall on held-out prompts, and it cannot hear a melody someone describes or hums.
- Renders share one GPU with the live demos: 3 per demo session, 3 per API key per day.
- C2PA credentials are signed by a development CA: valid signature, untrusted issuer in public validators.
- No copyright is claimed in the output; the certificate records checks and licence terms, not ownership.
Models and licences
Standard tier. Licence posture: community licence (check the terms).
- Prompt guard, code layer: gazetteer of well-known names, look-alike folding, cover/cloning/type-beat patterns, a genre and era allow-list; the certificate and refusal log (CPU)decosa-api music module (decosa_api/verticals/music)AGPL-3.0-or-later
- Prompt guard, typed judgment: names the category (artist, work, voice, label or none) with a calibrated probability; one call per briefQwen3.8-27B (NVFP4)Apache-2.0
- Music generation: songs with vocals and lyrics (default engine)MiniMax-Music3MiniMax-Music3 Community License
- Music generation: fast drafts and instrumentals (MIT engine)ACE-Step 1.5 turbo + 5Hz LM 1.7BMIT
- Similarity check (CPU): melody by chroma alignment across keys and tempos, sound by CLAP window embeddings, both normalised per referenceLAION CLAP larger_clap_music + librosa chroma (services/music_embed)Apache-2.0
All quality evidence
Every sourced number on the tool’s Stack tab, by tier. Some are proxies from another task; their labels say so.
Lite · MIT engine, about 15 GB of GPU (3)
- Prompt guard on held-out prompts (79: 30 safe, 26 overt, 23 sneaky): precision 100%, recall 98% (48 of 49), 0 of 30 safe prompts refuseddecosa-api docs/evals/music-gen-cleared.md, 2026-09-25; prompts hand-written for the eval
- Planted near-copies flagged (held-out test, 240 clips): 80.8% at 2.3% false alarms on unrelated tracks (2 of 88)decosa-api docs/evals/music-gen-cleared.md, 2026-09-25; threshold set on the dev split
- Music quality: not measured yet
Standard · the hosted demo, MiniMax-Music3 on a 96 GB card (3)
- Prompt guard on held-out prompts (79): precision 100%, recall 98%; code rules alone: recall 55%decosa-api docs/evals/music-gen-cleared.md, 2026-09-25
- Planted near-copies flagged, by change (held-out): pitch +2: 75%, pitch -1: 85%, tempo 0.92: 83%, tempo 1.08: 85%, pitch +1 and tempo 1.05: 85%, 12 s excerpt pitched -2: 73%decosa-api docs/evals/music-gen-cleared.md, 2026-09-25
- Music quality: not measured yet