The authors designed 40 controlled tasks, each pairing an initial image and generation prompt with predefined physical criteria.
Physical measurements expose persistent flaws in video world models
Across 40 controlled physics tasks, the strongest of eight video generators scored only 57.76 out of 100.
Top University
Mingju Gao · Qingle Liu · Yuzhao Peng · Xinjie Lin · Ziming Qin · Zheng Jiang · +5 more
Navers Lab · Einsia.AI · Peking University · Tsinghua University
Research Digest··2 min read
Gao and colleagues introduce a benchmark that tests whether generated videos obey measurable physical relationships rather than merely look plausible.
Why this paper
From Peking University and 3 others
In one line
A 40-task measurement-based benchmark for video world models finds persistent physical inconsistencies, with the best model scoring 57.76 out of 100.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§