minimax-h3

Use text-to-video with a concrete aspect ratio rather than adaptive, and write a prompt covering subject, action, camera path, environment, lighting, and audio intent. Keep the requested motion achievable within the 4-15 second clip duration.

These are provider-specific identifiers for the same Hailuo 3.0 family. The MiniMax v2 API uses MiniMax-H3, fal.ai uses hailuo-03, and Runway uses hailuo3. Do not substitute one provider's identifier into another API.

Yes, through the ComfyUI open-weights route using the official workflow exported in API format. It requires a local GPU plus the diffusion model, Qwen3-VL text encoder, video VAE, and audio VAE, and works offline once all models are installed.

Yes, first/last-frame animation is supported by providing both images and describing the motion between them in the prompt. Image-to-video requires only a first frame, while reference-to-video can combine images, videos, and audio.

The direct MiniMax v2 route outputs 2K with 4-15 second durations. Runway's Hailuo 3.0 route supports 768P or 2K with documented 5-15 second durations. Limits vary by provider, so check the chosen route before generating.