First test batch of 30 images. Rejected: many pages came out as multi-panel collages, faces changed from shot to shot, and long prompts were cut off. Round 2 uses single-panel prompts, short tags and approved reference portraits.



This preview uses PLACEHOLDER pictures for shots not generated yet with the real English voices. It shows pacing and sound only.
| Check | Result | Detail |
|---|---|---|
| duration within tolerance | PASS | 148.8s against target 180s (+-40%) |
| loudness near target | PASS | -15.5 LUFS against -16.0 |
| no black stretch over 1 s | PASS | 0 stretches |
| no freeze over 6 s | PASS | 0 stretches |
$0.00 spent of a $0.00 cap. This pilot uses free models only (Colab GPU and Piper), so cost is expected to stay at zero.
Send your verdict and edits to Claude in the session; they are applied to the shot list and re-rendered.