Oxford-led survey maps video world models to four operations

A 102-page survey posted to arXiv on 22 October proposes reading the whole field of video modeling through four verbs: reading, writing, sharing, and interacting. The paper, "Reading, Writing, Sharing, and Interacting with the World through Video", is led by Hao Fei at the University of Oxford with co-authors from NUS, NTU, and UC Merced, and carries the subtitle "A Survey Toward Unified Video World Models."

The pitch is a unifying frame. "Video foundation models have advanced rapidly in understanding and generating visual sequences, while video world models seek to capture the underlying dynamics of the world they depict," the authors write. "Despite their different objectives, both concern the relationship between video observations and an evolving world."

From there, the four operations. "Reading infers world properties from observed video; writing synthesizes video from specified or learned world conditions; sharing integrates understanding and generation through common modeling mechanisms; and interacting predicts and updates world states under actions and evolving observations." The authors argue the field is drifting away from clip-level recognition and perceptual fidelity toward "persistent state, controllable evolution, physical and causal validity, and evaluation grounded in interaction and downstream decisions."

The abstract names no models and publishes no benchmark numbers; the taxonomy is the product. A companion project page lists a public reading list, taxonomy, codes and data at vid-gen.github.io/video-world-model-survey.