film-generate-review
Installation
SKILL.md
Generate & Review (stage 6)
Why parallel
A 30 s 1080p take can take 5–20 minutes. Generation is also stochastic: the same prompt yields different eye-lines, different hand physics, a different third chair. The course rerolled serially and spent days. We fire four at once so that (a) the turn stays alive and (b) we see the distribution of what the prompt produces — which is the only way to tell "prompt problem" from "bad luck".
| Mode | Prompt files | Copies | Total | When |
|---|---|---|---|---|
lite |
1 | 1 | 1 | user explicitly asks |
standard (default) |
1 | 4 | 4 | normal |
max |
4 variants | 4 | 16 | user wants speed over credits, or the last standard batch was 4/4 wrong |
State the mode, count, model and expected wait before firing.
Scripts (film-pipeline/scripts/)
The helper below forwards images and parameters, not speech attachments. For voiced scenes, complete film-dialogue-voiceover, then use the configured audio-capable MCP/API and verify its submitted speech inputs. Never send an image-only batch as if it contained the approved voices.