phyground

A physics-grounded benchmark for video generation. Browse model outputs by physical law, compare side-by-side, and explore evaluator-by-dataset leaderboards.

Paper (coming soon) Leaderboard GitHub Dataset Model

Featured comparison — one prompt, 8 models

“A player hits the ball off the side of the mallet, resulting in an unexpected trajectory.”

Physical laws: collision impenetrability momentum

cosmos-predict2.5-14b Score: 2.33
cosmos-predict2.5-2b Score: 2.33
ltx-2-19b-dev Score: 1.00
ltx-2.3-22b-dev Score: 3.67
omniweaving Score: 3.67
veo-3.1 Score: 5.00
wan2.2-i2v-a14b Score: 2.67
wan2.2-ti2v-5b Score: 2.33

Open in compare view → More collision videos →

Sample random picks across laws & models — reload or shuffle for more

Browse

Leaderboard

51 evaluator×dataset slices across 8 models.

Video Gallery

Browse generated videos by physical law or by model.

Paper (coming soon)

Methodology, scoring schemas, and evaluation details — preprint pending.

About

Method overview, citation, and contact.