multi-clip-cinematic-generationlisted

# Generate one cinematic sequence, not unrelated clips This skill owns the decisions that must happen **before** an AI-generated follow-up clip exists. Read `minimax-h3-video`, `video-provider-capabilities`, and `fal-ai` before a paid call; endpoint controls are not interchangeable. Read `video-stitching` after clips are approved to plan and verify the edit. ## Choose the sequence topology first When the user has not set a per-clip duration, shot count, or beat boundary, use the smallest number of H3 generation boundaries that preserves the intended action. Those explicit user constraints win over cost and seam-count optimisation. Select and record one route for each seam: 1. `minimax/h3-max/text-to-video` for a prompt-only standalone opening shot with no reference or continuity obligation; 2. `minimax/h3-max/image-to-video` when an approved start image must control the opening composition; 3. `minimax/h3-max/reference-to-video` with identity-only references for a new shot, adding an accepted predecessor tail only when its state is needed; 4. a motivated editorial cut preserving the required identity and scene state; 5. an intentional discontinuity such as a time jump, location change, or montage beat. Never call separately generated clips a continuous take merely because a dissolve can join them. Reference-to-Video conditions a new successor; it does not append the predecessor or guarantee a frame-, mouth-, or audio-exact seam. Keep uninterrupted on-camera d