Give the model the goal and only the necessary visual anchors

Plan what the viewer must see and what must remain consistent. Separate must preserve, may change, and model discretion in the shot contract. Text direction is a valid starting point for clothing, locations, action, lighting and camera. An image is needed when it provides a specific control, not merely because a subject or location appears in the script.

Read minimax-h3-video for route controls and request handling, and multi-clip-cinematic-generation for the editorial handoff. Verify the live schema before using an input; endpoint support does not make that input necessary.

Start with the minimum sufficient references