HoYoverse-linked Anuttacon Unveils LPM 1.0: A Breakthrough in Video Character Performance Generation

Background and Release

Anuttacon, the AI company founded in Singapore by miHoYo co-founder Cai Haoyu, released the paper for LPM 1.0 (Large Performance Model), a video character performance generation model, on the arXiv preprint platform (reported as April 9, 2026). The model aims to turn a static image into a digital character capable of real-time conversation with fine-grained micro-expressions and body movements, achieving high consistency and long-duration stable video generation. The release is viewed as another step toward miHoYo's vision of "building a virtual world for a billion people by 2030."

Key Highlights

LPM 1.0 is designed for high-consistency video character performance generation, addressing the "performance trilemma" of expressiveness, real-time inference, and long-duration stability that traditional video models struggle to balance:

• Full-duplex real-time dialogue: The model simultaneously processes two audio streams — the user speaking (driving the character's listening reactions) and the AI character speaking (driving lip sync) — with low-latency streaming inference supporting continuous interaction of unlimited length. Official demos reportedly run 45+ minutes with stable character appearance and identity.
• Unlimited duration + extremely stable identity: Unlike traditional models that suffer character feature drift or collapse over long generations, LPM 1.0's online streaming architecture preserves identity consistency even across hours of continuous generation. Micro-expressions, eye movement, body rhythm, and emotional delivery are rendered with fine detail.
• Multimodal control: The model accepts image/reference video + audio + text prompts as input, with zero-shot generalization to realistic, 2D anime, 3D game styles, and even non-humanoid characters — no fine-tuning needed for new characters. Text controls actions, audio drives emotional expression, and the image defines character identity, enabling director-level control.
• Application scenarios: LPM 1.0 is positioned as a visual engine for conversational agents, virtual livestreaming, and game NPCs, turning a single image into a digital human that can talk, listen, and react in real time.

Technical Architecture and Training

LPM 1.0 uses a 17-billion-parameter Diffusion Transformer (DiT) architecture with multimodal conditioning for controllable, identity-consistent performance generation. The team built a human-centric multimodal dataset with strict curation and annotation of speaking-listening audio-video pairs, performance understanding, and identity-aware multi-reference extraction. Training proceeded in two stages: first a Base LPM (17B bidirectional DiT), then distilled into an Online LPM (a causal streaming generator) for low-latency, unlimited-length real-time interaction. The team also proposed LPM-Bench, a benchmark for systematically evaluating interactive character performance, on which LPM 1.0 reportedly achieves state-of-the-art (SOTA) results across all evaluation dimensions.

Background and miHoYo Connection

Anuttacon is an AGI company focused on interactive content and AGI products. It previously released the anime-style chat model "AnuNeko" and the AI-driven game *Whispers from the Star*. LPM 1.0 reflects Cai Haoyu's continued investment in the "AI + gaming" fusion. The model is currently for academic research only — no open source code, API, or commercial offering — but its results are seen as an important foundation for miHoYo's future AI applications in game NPCs and virtual worlds.

Community Reaction

• Positive: The model is praised for its clear advantages in long-duration consistency and expressiveness; Bilibili users called it the "most emotionally convincing" and noted it outperforms other video models (such as Kling).
• Neutral/skeptical: Some note that LPM 1.0 focuses narrowly on character performance rather than breadth, jokingly calling it miHoYo's consistent "digital waifu" route; others stress it remains at the paper stage with no usable product yet.

Overall, the technical quality is widely acknowledged, while the community takes a wait-and-see attitude toward practical deployment and commercialization.

Outlook

LPM 1.0 is a pragmatic, targeted step in the miHoYo/Cai Haoyu AI strategy — not chasing general-purpose scale but focusing on "character performance," the domain miHoYo knows best. If the technology lands in games like *Genshin Impact* or *Honkai* (or new projects), player interaction experiences could improve dramatically. For now, demos are impressive, but large-scale commercialization remains some distance away; follow the arXiv paper and project page for updates.

References: Anuttacon team's arXiv paper and project homepage; related tech media coverage.