In today's rapidly advancing field of artificial intelligence technology, MiniMax XiYu Technology released its latest video generation tool, Hailuo 02, on June 18th. This release not only represents a major breakthrough in the video generation domain but may also redefine how creators around the world produce their work.
One of the key features of Hailuo 02 is its ability to handle highly complex scenes, such as gymnastics that require precise expression. The introduction of this tool aims to help global creators better unleash their imagination, enhance video quality, and lower the barriers to creation. The MiniMax team has been dedicated to developing more efficient video generation model architectures, ultimately introducing the new Noise-aware Compute Redistribution (NCR) architecture.
With the NCR architecture, the training and inference efficiency of Hailuo 02 improved by 2.5 times, significantly enhancing the expressive power of the model with the same parameter volume. To achieve this goal, MiniMax increased the total number of parameters in the model threefold compared to the previous version, while the dataset size quadrupled. This means creators can generate high-quality video content at a lower cost.
Currently, Hailuo 02 offers three versions: 768p-6 seconds, 768p-10 seconds, and 1080p-6 seconds. Thanks to this new architecture, Hailuo 02 has a clear price advantage compared to similar products domestically and internationally. Additionally, during early testing phases, Hailuo 02 ranked second globally.
MiniMax stated that they will continue rapid updates across multiple domains in the future, including improving generation speed, refining content preference alignment, and achieving advanced functions such as image-to-video (I2V). The launch of Hailuo 02 marks a new starting point for video generation technology, making it a must-watch for all creators.
