Identity-Preserving Text-to-Video generation on IPT2V (test)
0.73FaceSim-ArcConsisID (Plan c)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ConsisID (Plan c)injection_strategy=high and low-frequency face information, embedding_location=attention block2024.11 | 0.73 | 0.75 | 36.77 | 127.42 | |
| IaD2026.06 | 0.67 | 0.71 | 28.75 | 114.67 | |
| ConsisID (Plan b)injection_strategy=only low-frequency face information2024.11 | 0.66 | 0.68 | 34.48 | 104.34 | |
| ConsisID (Plan d)injection_strategy=inclusion of facial key points2024.11 | 0.64 | 0.68 | 30.69 | 177.65 | |
| ConsisID (Plan e)injection_strategy=high and low-frequency face information, embedding_location=output block2024.11 | 0.62 | 0.66 | 33.61 | 164.15 | |
| ConsisID2026.06 | 0.57 | 0.54 | 27.15 | 144.35 | |
| ID-Animator2026.06 | 0.31 | 0.34 | 24.13 | 129.26 | |
| ConsisID (Plan a)injection_strategy=only high-frequency face information2024.11 | 0.05 | 0.05 | 34.86 | 269.88 |