ResearchBenchmarksScene Generation Evaluation on User Study n=90 trials per comparisonFollow3.3Loss RateSceneCode1.96410.9822029.018May 19, 2026Evaluation ResultsMethodMethodLinksLoss RateΔ PFΔ RealismSceneCodeGroup=C, Baseline=Layo...Group=C, Baseline=LayoutVLM2026.053.31.231.93SceneCodeGroup=B, Baseline=HSMGroup=B, Baseline=HSM2026.05200.660.27SceneCodeGroup=A, Baseline=Scen...Group=A, Baseline=SceneSmith2026.0536.70.140.34