miHoYo AI Paper Surfaces: Dynamic Character Animation Generation Gets 10x Faster

An AI paper recently surfaced by miHoYo is generating buzz in game technology circles. The paper proposes a diffusion-model-based method for dynamic character animation generation that can produce high-quality, coherent character animation from a text description or a simple motion sequence in an extremely short time. According to the paper’s data, the technology boosts generation efficiency by up to 10x compared with traditional animation pipelines.

The core of the technology is decomposing character motion into pose sequences, then using a diffusion model to progressively refine them until natural, fluid animation emerges. miHoYo demonstrates a range of scenarios in the paper, from “turning around while running” to “complex combat combos,” with animation quality approaching hand-tuned levels. Compared with the previously popular GAN and VAE approaches, diffusion models perform better in motion diversity and detail richness—and they require no large amounts of paired data.

For game developers, the breakthrough means the character animation production pipeline could undergo fundamental change. Traditional animation requires animators to adjust frame by frame or use expensive motion capture equipment—long cycles, high costs. miHoYo’s AI technology can dramatically shorten the time from idea to implementation, making it especially suited to high-frequency scenarios in open-world games such as real-time NPC reaction animation and player-customized moves. Take Genshin Impact: with hundreds of characters, AI-assisted animation generation could significantly reduce labor investment while giving each character a more distinctive movement style.

In terms of industry trends, miHoYo is not alone. In recent years, companies such as NVIDIA and Unity have all been positioning themselves in AI animation. NVIDIA’s Audio2Face technology generates facial animation from speech, while Unity’s Muse tools offer AI-driven animation generation services. What makes miHoYo’s paper distinctive is its focus on full-body dynamic control for game characters—including the coordination of limbs, torso, and fingers—which remains a hard problem in current AI animation.

The spread of AI animation still faces challenges, however. First, the “feel” and physical plausibility of generated motion still require human verification—especially in fighting or parkour games, where tiny motion deviations affect the player experience. Second, copyright ownership of AI-generated animation remains disputed; training on large amounts of public motion data could raise originality issues. In addition, real-world deployment requires deep integration with existing game engines (such as Unity and Unreal Engine), and miHoYo has not yet announced specific integration plans.

Looking ahead, miHoYo’s AI paper could become a catalyst for the democratization of game development. Small studios and even indie developers could use similar technology to produce high-quality animation at low cost, narrowing the content gap with the big players. At the same time, real-time AI animation generation will push NPC interaction from preset scripts toward dynamic responses, making player-world interaction more natural. As an industry leader, miHoYo is not only showing off its technical reserves with this move—it is signaling that AI will play an increasingly central role in game content production.

For players, AI animation technology may mean richer character moves and smoother cutscenes in the short term—and in the long term, open worlds that truly come alive. miHoYo’s paper has undoubtedly injected new imagination into AI game development.