In the current booming field of artificial intelligence, another domestic unicorn company, MiniMax, has announced its official entry into the fierce competition of video generation models. On August 31st, MiniMax successfully held the "MiniMax Link Partner Day" event in Shanghai, where founder Yan Junjie made a high-profile appearance and announced the company's latest development in video generation models and music models, marking the official entry of this low-key member of the "AI Six Dragons" into the new blue ocean of video generation.
During the event, Yan Junjie focused on introducing the video generation model named "video-1", which is known for its high compression rate, excellent text response capability, and diverse styles. It can generate native high-definition, high-frame-rate video content. Although specific technical details have not been fully disclosed, Yan Junjie revealed that video-1 currently supports text-based video functions and will be iteratively upgraded to image-based video, editing, and highly controllable advanced functions in the future, bringing users an unprecedented creative experience.
In the on-site experience session, reporters successfully generated a 6-second video with clear images and harmonious tones by simply inputting simple prompts, taking only 1-2 minutes, demonstrating the initial strength of the video-1 model. However, Yan Junjie also candidly pointed out that there is still room for improvement in the model's handling of facial details.
In the conference discussions, Yan Junjie delved into many non-consensus issues in the field of large-scale models, such as market positioning (2B vs 2C), geographical selection (domestic vs overseas), and the sustainability of scaling law. He emphasized that despite many uncertainties, video generation has become a widely recognized future development direction in the industry.
Since the beginning of this year, the field of video generation models has experienced explosive growth. From OpenAI's Sora to Shengshu Technology's Vidu, and then to Kuaishou, Luma AI, Runway, Alibaba DAMO Academy, Aishike Technology, and Zhipu, dozens of video generation models have emerged like mushrooms after rain in just a few months, marking the official entry of AI video generation technology into a historic development stage.
Yan Junjie stated that the reason MiniMax chose to layout video generation is because the information transmission in human society is increasingly relying on multimodal content. He pointed out that a large amount of information in daily life is transmitted through non-text forms such as voice and video, which requires large-scale model manufacturers to be able to output diverse forms of content to meet the wide range of user needs.
However, he also admitted that the challenges of video generation technology should not be underestimated. The current models still have many shortcomings in understanding physical rules and controlling the generation process, and the difficulty of processing video data is much higher than that of text, which places higher demands on infrastructure and research and development patience. Nevertheless, Yan Junjie is confident in MiniMax's technical strength and expressed that they will continue to invest resources to promote continuous breakthroughs in video generation technology.
With optimistic predictions from investment institutions such as Sequoia Capital about the future trends of generative AI, the full-scale outbreak in the field of video generation seems to be just around the corner. MiniMax's entry undoubtedly injects new vitality into this track and brings unlimited possibilities to the future production modes of film, animation, and short films.
