WanToFight: Real-Time Generative Game Engine for Multi-Player Combat Interaction
Summary
WanToFight is a generative game engine designed to simulate real-time, two-player The King of Fighters '97 (KOF '97) gameplay directly from keyboard input. This engine uniquely addresses the joint challenges of multi-player control, real-time inference, complex physical interaction, and adversarial gameplay, which prior generative game engines had not combined. It integrates three core components built on the Wan-1.3B video diffusion transformer: a streaming autoregressive generator with block-causal attention and a rolling KV cache; a visually grounded Player Association module; and a gated, locally causal keyboard injection module. A four-step DMD-distilled student, paired with a pruned VAE decoder, achieves 30FPS at 512x384 on a single NVIDIA RTX 5090 throughout a complete match.
Key takeaway
For AI Engineers developing real-time interactive generative models, WanToFight demonstrates a critical advancement in handling multi-player control and adversarial dynamics. You should examine its architecture, particularly the streaming autoregressive generator and the Player Association module, as a blueprint for achieving high-fidelity, low-latency generative experiences. This approach offers a path to overcome previous limitations in complex, real-time AI-driven simulations.
Key insights
WanToFight pioneers a generative game engine combining multi-player control, real-time inference, complex physical interaction, and adversarial gameplay.
Principles
- Generative engines can support complex multi-player combat.
- Video diffusion transformers enable real-time interactive AI.
- Curriculum learning aids complex behavior generation.
Method
It employs a streaming autoregressive generator, a Player Association module for character binding, and a keyboard injection module, trained with a single-player-to-full-gameplay curriculum.
In practice
- Simulate two-player combat from keyboard input.
- Achieve 30FPS real-time generative gameplay.
- Utilize a pruned VAE decoder for efficiency.
Topics
- Generative Game Engine
- Multi-Player AI
- Real-Time Inference
- Video Diffusion Transformer
- The King of Fighters '97
- NVIDIA RTX 5090
Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Engineer
Related on AIssential
See Counsel's argued verdicts on the open AI decisions leaders are weighing →
Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.