MiniMax has recently quietly launched its first large-scale video generation model and simultaneously released a 2-minute video titled "The Magic Coin" generated by this model. Although the company has not disclosed the specific parameters and technical details of the model, founder Yan Junjie stated in a media interview that the video generation effect is superior to Runway.
Yan Junjie revealed that the current release is only the first version of the model, and subsequent iterations will continue in areas such as data, algorithms, and usage details. In addition to the existing text-to-video function, future capabilities will include image-to-video and combined text-image-to-video generation. Regarding commercialization plans, Yan Junjie said they would consider them once the new version meets satisfactory conditions.
Compared to Kuaishou's Kelin, MiniMax's video generation model was launched a month or two later. Yan Junjie explained that this was because the team was addressing more challenging technical issues, particularly how to train content with higher computational power. He emphasized that MiniMax's core research approach is to pursue significant performance improvements, rather than just minor enhancements.
Image Source Note: The image was generated by AI, provided by the image licensing service provider Midjourney
Yan Junjie believes that the core motivation for developing video generation capabilities is to enhance user coverage and usage. He pointed out that the content humans consume daily is mainly in the form of text and video, so multi-modal content generation is an inevitable development direction.
However, large-scale video generation models face numerous challenges. Yan Junjie explained that the complexity of video generation is far greater than that of text, including the need to handle long contexts, huge storage requirements, and infrastructure upgrades.
Wei Weiye, the head of MiniMax's open platform, pointed out that the main challenges faced by current large models include inevitable hallucinations, high usage costs, and the development of multi-modal applications. He believes that as the cost of APIs further decreases, it will stimulate the emergence of more application scenarios.
In the face of many industry controversies, such as whether to focus on B2B or B2C, domestic or overseas markets, Yan Junjie expressed that MiniMax maintains an optimistic attitude towards technological progress, user engagement, and product iteration efficiency.