Chain of Event-Centric Causal Thought for Physically Plausible Video Generation

Zixuan Wang; Yixin Hu; Haolan Wang; Feng Chen; Yan Liu; Wen Li; Yinjie Lei

cs.CVMar 10, 2026

Chain of Event-Centric Causal Thought for Physically Plausible Video Generation

Authors: Zixuan Wang, Yixin Hu, Haolan Wang, Feng Chen, Yan Liu, Wen Li, Yinjie Lei

Organizations: Sichuan University · The University of Adelaide · Hong Kong Polytechnic University · University of Electronic Science and Technology of China

Abstract

Physically Plausible Video Generation (PPVG) has emerged as a promising avenue for modeling real-world physical phenomena. PPVG requires an understanding of commonsense knowledge, which remains a challenge for video diffusion models. Current approaches leverage commonsense reasoning capability of large language models to embed physical concepts into prompts. However, generation models often render physical phenomena as a single moment defined by prompts, due to the lack of conditioning mechanisms for modeling causal progression. In this paper, we view PPVG as generating a sequence of causally connected and dynamically evolving events. To realize this paradigm, we design two key modules: (1) Physics-driven Event Chain Reasoning. This module decomposes the physical phenomena described in prompts into multiple elementary event units, leveraging chain-of-thought reasoning. To mitigate causal ambiguity, we embed physical formulas as constraints to impose deterministic causal dependencies during reasoning. (2) Transition-aware Cross-modal Prompting (TCP). To maintain continuity between events, this module transforms causal event units into temporally aligned vision-language prompts. It summarizes discrete event descriptions to obtain causally consistent narratives, while progressively synthesizing visual keyframes of individual events by interactive editing. Comprehensive experiments on PhyGenBench and VideoPhy benchmarks demonstrate that our framework achieves superior performance in generating physically plausible videos across diverse physical domains. Code is available at https://github.com/ZixuanWang0525/CoECT.

Figures & tables

Figure 2: Overview of our PECR module ( Sec. 3.2 ). This module conceptualizes physical phenomena in user-provided descriptions as a series of causally ordered events governed by real-world physical formulas, where each event encompasses semantic descriptions and measurable physical parameters for key objects. This characterizes the underlying scene changes induced by such phenomena.

Figure 3: Overview of our TCP module ( Sec. 3.3 ). This module aims to generate semantic-visual prompts for each event based on its meta data. Semantic prompts are inferred by our proposed progressive narrative revision, serving as the guidance during denoising steps. Visual prompts are obtained by our proposed interactive keyframe synthesis, replacing original noise to provide physics-aware priors.

Figure 4: Visualization of physics-aware video generation results across four physical domains. Compared with baseline CogVideo-5B [ 38 ] , our approach yields causally coherent progressions of physical phenomena, e.g. , the glass ball sinks, the bottom shadow extends in the direction of the light, gradual melting of ice, fire spreading through the paper. All prompts are sourced from PhyGenBench [ 9 ] .

Video Foundation Model
Methods	Physical domains ( $\uparrow$ )				Avg.
Methods	Mechanics	Optics	Thermal	Material	Avg.
Lavie [ 40 ]	0.30	0.44	0.38	0.32	0.36
VideoCrafter v2.0 [ 41 ]	-	-	-	-	0.48
Open-Sora v1.2 [ 42 ]	0.43	0.50	0.34	0.37	0.44
Vchitect v2.0 [ 43 ]	0.41	0.56	0.44	0.37	0.45

Table 1: Performance comparison on PhyGenBench [ 9 ] across four physical domains. The best and second-best results are highlighted and underlined , respectively.

Video Foundation Model
Methods	Overall (%)			Solid-Solid (%)			Solid-Fluid (%)			Fluid-Fluid (%)
Methods	SA, PC	SA	PC	SA, PC	SA	PC	SA, PC	SA	PC	SA, PC	SA	PC
VideoCrafter2 [ 41 ]	19.0	48.5	34.6	4.9	31.5	23.8	27.4	57.5	41.8	32.7	69.1	43.6
LaVIE [ 40 ]	15.7	48.7	28.0	8.5	37.3	19.0	15.8	52.1	30.8	34.5	69.1	43.6
SVD-T2I2V [ 48 ]	11.9	42.4	30.8	4.2	25.9	27.3	17.1	52.7	32.9	18.2	58.2	34.5
OpenSora [ 42 ]	4.9	18.0	23.5	1.4	7.7	23.8	7.5	30.1	21.9	7.3	12.7	27.3

Table 2: Performance comparisons on VideoPhy [ 10 ] across various physical interactions between objects. The best and second-best results are highlighted and underlined , respectively.

Figure 5: Visualization of physics-aware video generation results across various physical interactions between objects. Compared with baseline CogVideo-5B [ 38 ] , our approach demonstrates clearer causal progression, e.g. , butter spread along the knife movement, continuous honey inflow with a rising level, and monotonic spring compression. All prompts are sourced from VideoPhy [ 10 ] .

Ablations of PECR module
Variant	Physical domains ( $\uparrow$ )				Avg.
Variant	Mechanics	Optics	Thermal	Material	Avg.
Ours	0.67	0.72	0.65	0.60	0.66
w/o PFG	0.63	0.69	0.61	0.53	0.62
w/o PPD	0.58	0.67	0.61	0.52	0.59
Ablations of TCP module

Table 3: Ablation analysis of PECR and TCP modules, including Physics Formula Grounding (PFG) and Physical Phenomena Decomposition (PPD) in PECR, and Progressive Narrative Revision (PNR) and Interactive Keyframe Synthesis (IRS) in TCP.

Figure 6: Ablation of the PECR module. Physical-driven CoT reasoning in PECR is crucial for modeling gradual changes in physical conditions and preserving causal consistency across events.

Figure 7: Ablation of visual prompts in TCP module. Physically consistent visual keyframe prompts are crucial, as unrealistic keyframes mislead the generator and lead to structurally implausible content.

Causal Video Generation

See all 21 papers

Physically Plausible Video Generation via Visual-Semantic Chain-of-Events Conditioning
Sep 1, 2026Zixuan Wang, Yixin Hu, Wen Li +4Physical Consistency in Video Generation Video Diffusion Models
College of Electronics and Information Engineering, Sichuan University, Chengdu, 610064, China · School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, 611731 · School of Computer Science, University of Adelaide, 5005, Adelaide, Australia +2
CausalMotion: Structured Physical Reasoning as Keyframe and Trajectory Guidance for Training-Free Video Generation
Jun 12, 2026Sihan Zhuang, Xinyuan Chen, Tianfan Xue +1Physical Consistency in Video Generation Diffusion Model Guidance
Shanghai Artificial Intelligence Laboratory · ShanghaiTech University · Work done during an internship at Shanghai AI Laboratory. +1
PhysVid: Physics Aware Local Conditioning for Generative Video Models
Mar 27, 2026Saurabh Pathak, Elahe Arani, Mykola Pechenizkiy +1Physical Consistency in Video Generation Video Generation
Eindhoven University of Technology

Sep 1, 2026J/K move · Enter open · S save

Physically Plausible Video Generation via Visual-Semantic Chain-of-Events Conditioning

Zixuan Wang, Yixin Hu, Wen Li, Feng Chen+3

College of Electronics and Information Engineering, Sichuan University, Chengdu, 610064, China · School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, 611731 · School of Computer Science, University of Adelaide, 5005, Adelaide, Australia+2

Physical Consistency in Video Generation Video Diffusion Models Causal Reasoning in Videos Controllable Video Generation Physical Reasoning in Video

Open paper AskPDF arXiv

Physically Plausible Video Generation (PPVG) seeks to synthesize videos consistent with physical principles, yet remains challenging due to underspecified natural language conditioning. Advanced chain-of-thought (CoT) frameworks augment prompts with physical knowledge. However, such prompts describe physical phenomena holistically, overlooking intermediate states and transition dynamics. In this paper, we reformulate PPVG as event-centric generation by representing physical evolution as a chain of causally connected and physically constrained events. Our framework comprises three key modules: (1) Physics-driven Event Chain Reasoning. This module decomposes physical phenomena into causally connected events represented by evolving scene graphs. Formula-derived physical quantities are bound to relevant objects and interactions, characterizing the direction and magnitude of each event transition. (2) Transition-aware Routed Keyframe Conditioning. This module routes each event to a specialized keyframe synthesis operator for appearance variation or object transformation. Consecutive keyframes are injected as residual guidance during denoising, enabling smooth visual transitions between event-boundary states. (3) Physics-injected Contrastive Semantic Guidance. This module constructs physics-informed positive and counterfactual negative prompts for classifier-free guidance, steering generation toward plausible dynamics and away from physics-violating counterparts. Experiments on PhyGenBench, VideoPhy, PhyWorldBench, and Physics-IQ demonstrate that our framework generates videos with superior physical plausibility across diverse domains.

Abstract

Causal Video Generation

Explore similar work

Physically Plausible Video Generation via Visual-Semantic Chain-of-Events Conditioning

CausalMotion: Structured Physical Reasoning as Keyframe and Trajectory Guidance for Training-Free Video Generation

PhysVid: Physics Aware Local Conditioning for Generative Video Models