Joint Audio-Video Generation on JavisBench
26.2Audio-Video Consistency (AV-IB)LTX-2+OmniNFT
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| LTX-2+OmniNFTSize=19B, Pipeline Approach=T2AV2026.05 | 26.2 | 3.326 | 5.715 | 0.261 | 0.189 | 31 | 44.5 | 25.7 | 22 | 0.269 | |
| LTX-2+ GDPOSize=19B, Pipeline Approach=T2AV2026.05 | 23.3 | 3.209 | 5.523 | 0.265 | 0.184 | 30.8 | 42.8 | 22.3 | 18.5 | 0.412 | |
| LTX-2Size=19B, Pipeline Approach=T2AV2026.05 | 23.2 | 2.038 | 5.197 | 0.272 | 0.17 | 31.1 | 41.2 | 22.3 | 19.2 | 0.569 | |
| MMAudioSize=0.1B, Pipeline Approach=T2V+V2A2026.05 | 19.8 | — | — | — | 0.16 | — | 40.7 | 18.2 | 15 | 0.849 | |
| JavisDiT++Size=2.1B, Pipeline Approach=T2AV2026.05 | 19.8 | 1.462 | 5.049 | 0.282 | 0.164 | 31.6 | 42.4 | 18.4 | 15.9 | 0.832 | |
| JavisDiTSize=3.1B, Pipeline Approach=T2AV2026.05 | 19.7 | 1.291 | 4.478 | 0.263 | 0.143 | 30.2 | 39.1 | 17.9 | 15.4 | 1.039 | |
| FoleyCrafterSize=1.2B, Pipeline Approach=T2V+V2A2026.05 | 19.3 | — | — | — | 0.149 | — | 38.3 | 18.6 | 15.1 | 0.952 | |
| See&HearSize=0.4B, Pipeline Approach=T2V+V2A2026.05 | 16 | — | — | — | 0.129 | — | 26.3 | 14.3 | 11.2 | 1.099 | |
| TempoTknSize=1.3B, Pipeline Approach=T2A+A2V2026.05 | 13.9 | — | — | 0.084 | — | 20.5 | — | 12.2 | 10.3 | 1.532 | |
| TPoSSize=1.0B, Pipeline Approach=T2A+A2V2026.05 | 12.4 | — | — | 0.201 | — | 22.9 | — | 12.9 | 9.5 | 1.493 | |
| ReWaSSize=0.6B, Pipeline Approach=T2V+V2A2026.05 | 11 | — | — | — | 0.123 | — | 28 | 10.4 | 7.9 | 1.071 | |
| UniVerse-1Size=6.4B, Pipeline Approach=T2AV2026.05 | 10.4 | 1.357 | 4.839 | 0.272 | 0.111 | 30.9 | 24.5 | 9.8 | 7.7 | 0.929 |