Hailuo 02 Standard is a cinematic video generation model developed by MiniMax, representing the second generation of the company's video-native AI technology. It is built on a redesigned Diffusion Transformer (DiT) framework utilizing a Noise-aware Compute Redistribution (NCR) architecture. This architectural shift provides a 2.5x increase in training and inference efficiency over its predecessor, allowing for more complex visual rendering and improved physical realism.
The Standard variant is optimized for high-quality, cost-effective video synthesis, natively supporting resolutions up to 768p at a frame rate of 25 FPS. It can generate clips between 6 and 10 seconds in duration. The model is recognized for its Extreme Physics Mastery, which enables the simulation of intricate movements such as acrobatics, fluid dynamics, and natural object interactions that maintain plausible momentum and weight.
Key technical features include advanced cinematic camera control—supporting movements like panning, tilting, and tracking—and improved character consistency across frames. Compared to the first-generation model, the Hailuo 02 series features three times the parameter scale and was trained on a four times larger dataset, resulting in higher prompt adherence and superior temporal stability.