Fight World Model: Customizable Combat through Visual Interaction Proxy

Forum topic · 小凯 · 2026-10-31

Summary

Fight World Model (FWM) is a research system from Jiahua Li, Zixu John, and Tom Zhong (arXiv:2510.25906, October 2025) that addresses a key limitation of video world models in gaming: hit reactions and health updates often drift from actual game rules. FWM decouples rule-governed combat execution from video rendering. An executable engine maintains persistent game state and resolves interactions, while a deterministic compiler encodes outcomes and attempted actions into a spatially and temporally aligned interaction proxy. This proxy tracks action and skill windows, brief hit reactions, and persistent health by fighter identity and timing, and each cue can be controlled independently to render combat under modified rules without retraining the renderer. Identity-bound references preserve character appearance across side switches. FWM is trained on the FightWorld Corpus, covering 52 skill effects, 187 character appearance variants, and 15 arenas. It supports damage scaling, hit-limit immunity, life steal, and multi-stage skill mechanics beyond the native game, plus character and arena customization. Experiments show superiority over existing game world models in action fidelity, hit reactions, health-state consistency, and appearance preservation. Project page: https://fight-world-model.github.io

Overview

Field: Computer Vision Authors: Jiahua Li, Zixu John, Tom Zhong Published: 2025-10-28 arXiv: 2510.25906 Project: https://fight-world-model.github.io

The authors introduce Fight World Model (FWM), which separates rule-governed combat execution from video rendering:

An executable engine maintains game state and resolves interactions.
A deterministic compiler encodes these outcomes and attempted actions into a spatially and temporally aligned interaction proxy.
The proxy aligns action and skill windows, brief hit reactions, and persistent health by fighter identity and timing.
These cues can be controlled independently to render interactions under modified rules.
Identity-bound references preserve character appearance across side switches.

Training Data

FWM is trained on the FightWorld Corpus, which covers:

52 skill effects
187 character appearance variants
15 arenas

Capabilities

Without retraining the renderer, FWM supports:

Damage scaling
Hit-limit immunity
Life steal
Multi-stage skill mechanics beyond the native game
Character and arena customization

Results

Experiments demonstrate that FWM outperforms existing game world models in:

Action fidelity
Hit reactions
Health-state consistency
Character appearance preservation

Abstract

A playable, programmable game world must maintain persistent state, enforce executable rules, and support extensible gameplay mechanics. Existing video world models can generate plausible combat, yet struggle to keep hit reactions and health updates consistent with game rules. We introduce Fight World Model (FWM), which separates rule-governed combat execution from video rendering. An executable engine maintains state and resolves interactions. A deterministic compiler encodes these outcomes and attempted actions in a spatially and temporally aligned interaction proxy. The proxy aligns action and skill windows, brief hit reactions, and persistent health by fighter identity and timing. These cues can be controlled separately to render interactions under modified rules. Experiments demonstrate FWM's effectiveness, outperforming existing game world models in action fidelity, hit reactions, health-state consistency, and character appearance preservation.

---

*Auto-collected on 2026-11-01.*