I just saw interesting news about AI character-rendering development launched by Anuttacon, an AI company founded by Tsai Haoyu, the founder of miHoYo.

They recently released LPM 1.0, a video rendering model for characters that is quite impressive. This system allows characters to speak, sing, listen, and react all at the same time in real time. What’s more, the movements of the mouth and facial expressions, as well as body timing and gestures, look extremely realistic—not at all like something artificial.

From a technical perspective, they used a base model with 17 billion parameters and adapted it into a low-latency streaming version. The good part is that it supports a wide range of character styles, from realistic appearances and 2D animations to 3D game characters, and even non-human creatures—without any additional customization.

What I noticed is that Anuttacon has also introduced the LPM-Bench evaluation standard. Compared with other models on the market, it achieves the best results across many metrics. The key difference is that LPM 1.0 can generate content for an unlimited duration, unlike other options that are restricted to only 30 seconds.

Their intended applications are fairly clear: NPC conversation agents in games and virtual live broadcasts are highly promising use cases, especially for miHoYo and other game companies.

However, it’s important to note that Anuttacon explicitly states that this is for academic exchange purposes only. There are currently no plans to open-source it or provide APIs to the public. Most of the community believes the main purpose is to attract capable AI talent.

Looking back, LPM 1.0 is the next step in Anuttacon’s AI roadmap for game engines and intelligent NPCs. It’s also another outcome of miHoYo’s AI development, from the chat tool AnuNeko to the capabilities of the base model. It seems that Tsai Haoyu and his team are building something significant in the field of AI character technology.