Hailuo 02

Choose the Fast tier for rapid storyboard sketches and layout prototyping at low compute cost, switch to Standard for balanced 768p cinematic motion across 6- or 10-second clips, and deploy Pro when you need movie-grade visual fidelity and maximum fine-detail preservation on hero visual effects plates. Fast operates exclusively in Image-to-Video mode with draft-scale output, whereas Standard and Pro support both Text-to-Video and Image-to-Video generation.

Select Image-to-Video when you have a pre-composed visual layout or concept still and need to preserve specific character features, materials, and framing. Text-to-Video is best suited for establishing shots, high-concept world-building, and dynamic atmospheric scenes where you want the model to generate composition, environment, and choreography directly from a written description.

Set a starting image in the First Frame slot and a matching target image in the Last Frame slot to generate continuous video interpolation between both states. Hailuo 02 computes realistic physical motion and camera transitions to bridge the two compositions cleanly, provided both frames share consistent lighting, subjects, and environmental structure.

Hailuo 02 excels at high-momentum physical acrobatics, fluid dynamics like water and smoke simulation, cinematic camera moves such as dolly-zooms and orbital pans, and temporal character locking. It struggles with fine typographic rendering, multi-scene narrative transitions within a single prompt, flat 2D motion graphics, and photorealistic human selfies which can trigger strict safety filters.

Hailuo 02 was developed by MiniMax and released in June 2025, with expanded model variants rolled out in August 2025. It is built on a Mixture-of-Experts (MoE) architecture with Noise-aware Compute Redistribution (NCR) to optimize diffusion compute efficiency.