Multi-Subject Video Generation on VLM-based Gemma-31B benchmark leaderboard · SOTA2 Research

Multi-Subject Video Generation on VLM-based Gemma-31B

4.3Action ScoreAura

Evaluation Results

Method	Links
Aura2026.07		4.3	5	4.85	4.35	3.889	4.981
Wan2.72026.07		4.25	4.9	5	4.25	4.911	4.911
HuMo2026.07		4.05	4.9	5	4.05	4.163	4.939
MAGREF2026.07		3.7	4.65	5	3.4	4.192	4.577
Kaleido2026.07		3.25	4.6	4.55	2.9	3.68	4.84
RefAlign2026.07		3.25	4.5	4.65	3.6	3.625	4.393