Creative Claw — MiniMax H3 Max

Use video/minimax-h3-max for fast, aesthetically strong 480P or 768P video with native synchronized audio, optional first/last frames, and multimodal references. Use video/minimax-h3-max-turbo only when the user prioritizes lower latency and cost and does not need reference video or reference audio.

Do not invent an h3-max-lite model ID. The current faster lightweight route is video/minimax-h3-max-turbo.

Core workflow

1. Define one shot: duration, ratio, subject, action, camera, audio, and continuity anchors.
2. Search or import source assets.
3. Generate and approve a clean storyboard frame with image/nano-banana-2. For a controlled transition, create both opening and ending frames before video generation.
4. Call get_model_params({ model: "video/minimax-h3-max" }) immediately before generation.
5. Choose text, first-frame, first-to-last, or reference mode deliberately.
6. Assign every reference a role using H3 Max's one-based Image 1, Video 1, and Audio 1 language.
7. Set agentic_prompting: false for exact reference labels, dialogue, timecodes, or locked prompt structure.
8. Generate at 480P for cheap motion tests or 768P for the preferred final H3 Max result.
9. Inspect visible motion and synchronized audio before reuse.

Choose the route