Text-to-Video Generation on CountBench benchmark leaderboard · SOTA2 Research

Text-to-Video Generation on CountBench

0.591Counting AccuracyWan2.1-14B + NUMINA

Evaluation Results

Method
Wan2.1-14B + NUMINA2026.04	0.591	0.84	0.344
Wan2.1-14B + Prompt enhancement2026.04	0.569	0.843	0.34
Wan2.1-14B + Seed search2026.04	0.561	0.835	0.34
Wan2.1-14B2026.04	0.536	0.833	0.342
Wan2.2-5B + NUMINA2026.04	0.527	0.85	0.347
Wan2.1-1.3B + NUMINA2026.04	0.497	0.834	0.356
Wan2.2-5B + Prompt enhancement2026.04	0.49	0.843	0.342
Wan2.2-5B + Seed search2026.04	0.488	0.847	0.341
Wan2.2-5B2026.04	0.478	0.85	0.343
Wan2.1-1.3B + Prompt enhancement2026.04	0.472	0.821	0.337
Wan2.1-1.3B + Seed search2026.04	0.455	0.823	0.346
NUMINA2026.04	0.444	0.802	0.354
CogVideoX-5B + Seed search2026.04	0.427	0.783	0.348
CogVideoX-5B + Prompt enhancement2026.04	0.425	0.79	0.345
Wan2.1-1.3B2026.04	0.423	0.812	0.339
CogVideoX-5B2026.04	0.402	0.781	0.348