Li et al.
Video generation models often fail to generate causally coherent physical event chains
A new benchmark reveals that current image-to-video models struggle to faithfully realize how events propagate through physical interactions, even when given detailed prompts.
Top University
Jialiang Li · Zhengning Xu · Zongyang Ma · Shaobo Cui
Shanghai Jiao Tong University · Xi'an Jiaotong-Liverpool University
Research Digest··3 min read
Thread:Video Physical Consistency
The authors introduce CausalCascade, a simulator-grounded benchmark of 50 Blender scenes with structured event graphs and 2,832 controlled counterfactual variants.
Why this paper
From Shanghai Jiao Tong University and Xi'an Jiaotong-Liverpool University · Part of Video Physical Consistency, now 3 papers
In one line
Current video models fail to generate causally coherent physical cascades, even with explicit event prompts.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§