Fluid-Gen-Zero Benchmark
Benchmark for fluid-object interaction video generation
A benchmark of 43 curated fluid-object interaction samples with 401-frame simulation rollouts, 3D trajectories, masks, and optical flow across 10 fluid categories.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info fluid-gen-zero-benchmark
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
6 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| WonderPlay | 0.4 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| PhysGen3D | 0.4 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| MotionCraft | 0.4 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| VACE | 0.5 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| Tora | 0.6 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| WanMove | 0.6 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
The benchmark comprises 43 curated samples featuring fluid-object interactions, including 401-frame simulation rollouts, 3D particle trajectories, object/fluid masks, and projected optical flow across 10 primary fluid categories.
It evaluates physics-aware video generation by measuring simulation alignment via object trajectory error and fluid fEPE, where Fluid-Gen-Zero reduced errors by 26.7%–81.5% and 67.9%–84.0% respectively.