01 / Written scene
A scene needs beats before it needs frames.
Vivid begins by turning a written idea into a small ordered plan: world, person, action, and the next image that has to earn its place.
- SceneA ceramic artist works through one quiet workshop moment.
- CharacterOne neutral anchor defines the person who must persist.
- ShotsFour ordered beats change framing and action without treating any one image as enough.
- ReviewThe sequence is judged together before a profile can be promoted.
02 / Character anchor
The person is the master pattern.
A neutral reference gives the sequence a person to keep, without making an earlier rendered pose the rule for every following shot.

03 / Reference bindings
The plan and anchor become the loom's threads.
The visible threads stand for planned continuity bindings. They are decorative movement; the ordinary narration carries the real boundary.
04 / Contact sheet
Four frames make one question visible.
These are real frames from the cleared Kyoto pottery evaluation sequence. Together they are more useful than one showcase image.




05 / Rejected speedup
Faster was not good enough to ship.
The faster Turbo profile stayed rejected after visual review found quality regressions. The baseline remained the production choice.
Vivid's accepted base-model evaluation covers 12 fixed scenes and 42 generated shots, with identity 0.3528, adherence 0.8437, and realism 0.8562.
42 generated shots across 12 scripted scenesA Vivid Turbo evaluation was 5.90 times faster across 42 shots, but it was rejected because identity and text-scene quality regressed.
42 shots across 12 scripted scenes06 / Evidence boundary
A missing visual stays missing.
The documented pose-copying finding is not reenacted here because the publishable before-and-after comparison is unavailable.
Visual comparison unavailable
The verified pose-copying finding has no cleared visual comparison for this page, so it is stated here without an illustrated substitute.
07 / Current boundary
A promising sequence is not a solved product.
The sequence is portfolio-cleared evidence. Commercial product-model licensing remains unresolved, and no claimed capability goes beyond the audited record.
Portfolio evidence, not a licence claim
The sequence is cleared for this portfolio as evaluation evidence; Vivid's commercial product-model licence is still unresolved.
08 / Deep-dive handoff
The loom stops. The case study keeps the record.
The paper case study carries the architecture, measured baseline, rejected Turbo result, failures, research and future work.
Read the full case study 