Personalized Audiovisual Generation on Identity-Referenced Audiovisual Generation Dataset S2 (test)
241.9FVDMAVIN
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MAVINPipeline Architecture=Joint, Reference Modality=Identity-Referenced AV Generation2026.06 | 241.9 | 6.9 | 0.2463 | 0.2391 | 0.049 | 6.057 | 0.266 | 0.793 | 0.9682 | 0.9653 | 0.6397 | 0.7996 | 0.9886 | 0.6259 | 0.4716 | |
| VACEPipeline Architecture=Cascaded, Reference Modality=Image-Referenced Video Generation, Downstream Model Pairing=V2A (Hunyuan-Foley [54])2026.06 | 253.4 | 12.3 | 0.2296 | 0.2293 | — | 2.513 | 0.212 | — | 0.9677 | 0.9574 | 0.5834 | 0.6997 | — | 0.6049 | — | |
| PhantomPipeline Architecture=Cascaded, Reference Modality=Image-Referenced Video Generation, Downstream Model Pairing=V2A (Hunyuan-Foley [54])2026.06 | 275.3 | 11.7 | 0.2318 | 0.2189 | — | 2.396 | 0.202 | — | 0.9663 | 0.9546 | 0.6026 | 0.7149 | — | 0.6285 | — | |
| CosyVoicePipeline Architecture=Cascaded, Reference Modality=Audio-Referenced Speech Generation, Downstream Model Pairing=A2V (MTV [70])2026.06 | 324.2 | 7.2 | 0.2014 | 0.2205 | 0.053 | 3.821 | 0.235 | — | 0.9583 | 0.9584 | 0.5739 | 0.7572 | — | — | 0.5002 | |
| F5-TTSPipeline Architecture=Cascaded, Reference Modality=Audio-Referenced Speech Generation, Downstream Model Pairing=A2V (MTV [70])2026.06 | 336.4 | 7.5 | 0.2099 | 0.2186 | 0.046 | 3.804 | 0.229 | — | 0.9591 | 0.9589 | 0.5813 | 0.7669 | — | — | 0.561 |