WanToFight generates game visuals based on real-time keyboard inputs, enabling accurate two-player control and complex character interactions.
We present WanToFight, a generative game engine that simulates real-time, two-player The King of Fighters '97 (KOF '97) gameplay from keyboard input. Prior generative game engines target either single-player first-person settings or non-real-time cooperative scenarios; multi-player control, real-time inference, complex physical interaction, and adversarial gameplay have not been jointly addressed. WanToFight closes this gap with three 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 that binds each player's keyboard signal to a character identity; and a gated, locally causal keyboard injection module trained with a single-player-to-full-gameplay curriculum. A four-step DMD-distilled student paired with a pruned VAE decoder sustains 30 FPS at 512×384 on a single NVIDIA RTX 5090 over the duration of a complete match. To our knowledge, WanToFight is the first generative game engine to combine multi-player control, real-time inference, complex physical interaction, and adversarial gameplay in one system.
Overview of WanToFight. The model is an autoregressive video diffusion engine that generates gameplay one chunk at a time. Given an initial frame and a stream of keyboard inputs, the model produces the next chunk by conditioning on the previously generated chunk through cached attention states. The Player Association module binds each player's keyboard signal to the corresponding character through reference-image-driven attention, and the Window Attention mechanism restricts each video latent to attend only to nearby keyboard latents, enforcing local causality of control.
@article{hu2026wantofight,
title={WanToFight: Real-Time Generative Game Engine for Multi-Player Combat Interaction},
author={Li Hu and Guangyuan Wang and Peng Zhang and Bang Zhang},
journal={arXiv preprint arXiv:2607.12592},
year={2026}
}