WorldExam is introduced, a hierarchical diagnostic benchmark spanning four levels: Visual Quality, Control Adherence, Spatial Consistency, and World Reactivity, which supports unified evaluation of camera-, action-, and language-driven model paradigms.
Abstract
Controllable video generation models are increasingly being developed as world models. Accordingly, evaluating them in this role extends beyond the apparent appearance of generated videos to the inherent reactivity of the worlds they depict: the ability to infer from the scene state how the world should react and to generate plausible consequences not explicitly described in the input. Yet existing benchmarks mainly assess visual quality or explicit instruction fulfillment by checking whether requested actions and interaction outcomes are realized, leaving inherent reactivity underexamined. We introduce WorldExam, a hierarchical diagnostic benchmark spanning four levels: Visual Quality, Control Adherence, Spatial Consistency, and World Reactivity. It comprises 1,474 cases across eight dedicated tasks and supports unified evaluation of camera-, action-, and language-driven model paradigms. The World Reactivity level evaluates scene-conditioned reactions and goal-directed behaviors beyond what is explicitly specified in the input. Evaluation of 20 representative models reveals a clear capability split. Camera-driven models excel at camera control, but their interfaces do not support dynamic interaction; action-driven models control subjects more precisely but often leave the world unresponsive; and language-driven models perform better on interaction but follow complex controls less faithfully. No model combines broad task coverage with consistently strong performance, showing that high visual quality and explicit instruction fulfillment do not guarantee inherent reactivity.
This work introduces PROWBench, comprising 170 programmatically constructed episodes and 600 proxy videos covering diverse scenes and interactions, and evaluates entity control, long-horizon memory, and adherence to the prescribed timeline and the visual realization of timestamped engine-recorded events.
Zheng-Hui Huang, Gui-Xu Lin, Yu-Ju Tsai et al.· 3 citations
World in World is presented, a training-free inference-time interface that converts heterogeneous control evidence into camera- and time-labelled clean visual states, which are read through the native self attention of a frozen causal video model.
Chen-Xi Song, Yan-Ming Yang, Chi Zhang· 1 citation
OPIS is introduced, an input-grounded benchmark that strictly anchors the assessment to a fixed set of object instances from the initial observation for evaluating multi-object memory in video world models and demonstrates that preserving the particular object instances in the input is considerably harder than generati...
Hao Wang, Tao Yu, Liu-Zhou Zhang et al.· arXiv.org· 1 citation
The Programmable World Model is introduced, a framework that decouples world-state evolution from visual observation generation and demonstrates the effectiveness of separating explicit state evolution from generative rendering for building persistent, programmable worlds.
Zheng-Hui Huang, Gui-Xu Lin, Jia-Cheng Lin et al.· 20 citations· ⚡3
This work formalizes probabilistic alignment as a distributional criterion for world models and introduces PAWBench, a benchmark for evaluating video generators as stochastic samplers of world dynamics, and introduces PAWEval, an outcome-level protocol that converts repeated video rollouts into empirical distributions...
Yuan-Dong Pu, Le Zhuo, Sayak Paul et al.· 0 citations
HappyWorld-Bench is introduced, a comprehensive benchmark that evaluates whether generated worlds remain reliable as agents interact with them, and highlights the need to evaluate world models not only by visual quality, but also by state consistency and the correctness of their responses to actions and interventions.