miHoYo AI Paper Surfaces: Anti-Entropy Team’s New Method Supercharges Character Animation

An AI paper from miHoYo’s Anti-Entropy team has recently drawn wide attention across the games and AI fields. Titled Prove You’re Human, the paper lays out a novel character animation generation method called Mesa, designed to tackle the high cost and low efficiency of animation production in game development today.

The heart of Mesa is an elegant combination of diffusion models and motion matching. Traditional diffusion models can generate high-quality animation, but their computational cost makes them ill-suited to real-time games. Motion matching, meanwhile, is highly efficient but constrained by the size of its prerecorded animation library. Mesa uses the diffusion model to generate a large pool of candidate animations, then lets a motion matching system rapidly select and blend them—dramatically boosting generation efficiency while preserving animation quality.

The breakthrough carries profound implications for the game industry, especially for open-world development. Take miHoYo’s own Genshin Impact: its characters already move with considerable richness, but more natural NPC interactions or more complex dynamic scenes still demand enormous amounts of hand-tuning. Mesa could compress animation production cycles severalfold, letting developers focus on the creative work itself.

Notably, miHoYo is far from the only company exploring AI animation. Epic Games’ MetaHuman Animator and Unity’s Muse platform are both trying to streamline character animation with AI. But miHoYo’s Mesa paper demonstrates a complete loop from generation to matching, with a technical approach far closer to how game engines actually work.

In terms of industry impact, Mesa’s debut could accelerate the adoption of AI in game animation. On one hand, small and mid-sized teams could use similar techniques to lower the animation barrier and craft more vivid characters; on the other, large studios could leverage it to build vaster open worlds—even achieving “infinite animation” effects. For traditional animators, though, it poses a new challenge: learning to collaborate with AI rather than being displaced by it.

In the future, as technologies like Mesa mature, we may see game character animation shift from painstaking hand-crafting to a new model of “AI generation plus human refinement.” Players will experience smoother, more natural character motion, while developers free up resources to polish gameplay and narrative. With this reveal, miHoYo has pointed the entire industry in a clear direction.