Our story film

Eval results

Not held outRun 29 Sep 2026Eval write-up (decosa-api, access required)

• Planned cuts found on the beat in the exports30 / 30syntheticn = 303 storyboard renders of the sample song; median 5.2 ms, max 16.0 ms from the beat grid (half a frame at 30 fps is 16.7 ms).
• Storyboard frames flagged by the frame safety check0 / 33syntheticn = 33Sample performers are synthetic adults; the check fails closed.
• Storyboard video done (p50)142.5 ssyntheticn = 3122-275 s (p50 142.5 s); first picture drawn after 15-17 s; shared GPU.
• Cost per storyboard video$0.040-0.074syntheticn = 3Text model at list price plus GPU time at $1.32/h.
• MiniMax H3 time per moving shotabout 80 s (two references)syntheticn = 11Sketch tier 864x480, fp8, one 96 GB card, peak 49.6-52 GiB; recorded in a GPU window on 29 Sep.
• MiniMax H3 cost per shot$0.022-0.029syntheticn = 27GPU time at $1.32/h; 27 shots including 6 re-rolls.
• Face likeness to the consent clip (ArcFace cosine, storyboard / H3)0.43 / 0.40syntheticInternal QC with a non-commercial model, not shipped. H3 shots were often wide or turned away (faces found in 15 of 80 sampled frames); frontal re-rolls reached 0.51.
• Moving shots that kept the face (likeness gate)5 / 11syntheticn = 11One middle frame per moving shot checked against each person's reference; the rest fell back to the still. Some wrong faces still pass.
• Blind partner: gives it / would pay (storyboard; moving)yes after one fix, $29; no, $0syntheticn = 1An Opus sub-agent buying an anniversary gift. The fix it asked for (his partner's face changed in one shot) is why both partners are now checked.

Nightly smoke check

Loading the nightly status…

Result

pass

Run

29 Sep 2026

Latency, this run

n/a

p50 over passed runs

143 s

Receipts

5

Model calls

n/a

Tokens

n/a

Cost per run

$0.041

Self-host verification

Not yet verified on a fresh self-host setup.

Models and licences

Standard tier. Licence posture: permissive (Apache, MIT or BSD).

• Consent read-back: hears whether the clip says the sentence and its three fresh wordsQwen3-ASR-1.7B (language pack speech service)Apache-2.0
• Face detection only (boxes): exactly one face, present and moving through the clip; its three sharpest frames become the only face referencesUltra-Light-Fast-Generic-Face-Detector-1MB (version-RFB-320)MIT
• Adult check on the consent frames (vision), the shot list per song section from your three memories, and the frame safety and likeness checksQwen3.8-27B (NVFP4)Apache-2.0
• Beat grid, bars and sections of the song (CPU; the music-video studio's analyzer): cuts land on bar downbeatsdecosa-mvideo-analyze (services/mvideo)AGPL-3.0-or-later (decosa-api)
• Storyboard: one still per shot drawn from the consent-clip references, then a camera move (push, pan, drift)FLUX.2 klein 4BApache-2.0
• The edit (CPU): shots cut on the bar downbeats at 30 fps, the AI video label on every frame, an end card crediting the music, exports in 16:9, 9:16 and a Spotify Canvas loop; cut timing measured back from the pixels; C2PA per filedecosa-api starring module + FFmpeg + c2pa-pythonAGPL-3.0-or-later

All quality evidence

Every sourced number on the tool’s Stack tab, by tier. Some are proxies from another task; their labels say so.