MiniMax Unveils Its First AI HD Video Generation Model,

TMTPOST--Chinese AI startup MiniMax introduced its first AI HD video generation model, abab-video-1, at the inaugural MiniMax Link Partner Day on August 31.

The company's founder and CEO, Yan Junjie, presented the technology, describing it as a significant step forward in AI-driven video creation.

Users can now experience this technology on the Conch AI website, where they can generate short videos featuring text prompts. The model supports native 1280x720 resolution at 25 frames per second, offering cinematic camera movements. Currently, the maximum video length is six seconds, with plans to extend it to ten seconds in the future. According to early reviews, the quality of these AI-generated videos rivals that of existing models such as Sora and Runway.

Yan also shared that MiniMax now facilitates over 3 billion global interactions daily, processing more than 3 trillion tokens—the highest in the industry. He revealed that the total daily interaction time on MiniMax is now at 53% of ChatGPT's, and the gap is narrowing.

In a discussion with reporters after the event, Yan addressed competition with larger companies. He acknowledged that competition is inevitable and essential for industry growth. "Competition drives demand and generates significant societal value. If we cannot win in this competition, then we deserve to be eliminated. There is no other choice for us as a small startup," Yan said.

He emphasized that technology development is a "luxury" that requires immense resources and comes with a high risk of failure. However, Yan remains optimistic, stating that breakthroughs in technology can lead to widespread user recognition and sustained growth. He believes that MiniMax's success lies in the continuous improvement of its technology and closer collaboration with users.

Yan also discussed MiniMax's approach to research and development, highlighting the importance of speed. "With similar algorithms, the faster model will always benefit more from data growth, leading to better overall performance. In this sense, speed is crucial," he said.

Looking to the future, Yan noted that MiniMax aims to address three key challenges in AI model development: reducing error rates, enabling infinitely long inputs and outputs, and supporting multi-modal capabilities. The company envisions a future where AI-generated content becomes an integral part of a "content community," offering users more tools to create stories and interact with the world.

Yan acknowledged that MiniMax's video technology launch came a few months later than competitors like Runway, but he emphasized that the team focused on overcoming the technical challenge of training AI on high-complexity video data. As a result, the company's new algorithms have significantly improved video quality by reducing complexity and increasing compression rates.

The CEO also revealed that MiniMax plans to update its video generation technology regularly, with potential future features including video generation from images and text, enhanced editing capabilities, and greater control over the final output. However, Yan stressed that the current focus is not on commercialization but on making the technology widely accessible.

Regarding business-to-business (B2B) applications, Yan mentioned that MiniMax's open platform now has over 30,000 developers. In terms of monetization, he expressed optimism about the growing demand for AI models in both domestic and international markets. He pointed out that while Chinese companies have made significant strides in AI, particularly in non-English languages, they are now increasingly competitive on the global stage.

"Competition is unavoidable, so we must strive to be the best," Yan concluded, highlighting MiniMax's commitment to achieving top-tier AI performance on a global scale.

As MiniMax continues to innovate, the company remains focused on pushing the boundaries of artificial intelligence and bringing intelligence to everyone through collaboration and cutting-edge technology.