Image Generation on ImageNet-1K 256
3.79FIDTeacher velocity network + 3-channel CFG
Evaluation results
| Method | Links | |
|---|---|---|
| Teacher velocity network + 3-channel CFGInitialization=flow-matching, Teacher Sampling Steps=250-step Euler sampling, CFG=3-channel2026.05 | 3.79 | |
| FSF-DMDInitialization=flow-matching2026.05 | 4.7 | |
| DMD2 + LCDInitialization=flow-matching, TTUR=1:52026.05 | 4.9 | |
| Consistency distillationInitialization=flow-matching2026.05 | 5.53 | |
| DMD2Initialization=flow-matching, TTUR=1:52026.05 | 5.53 | |
| Teacher velocity networkInitialization=flow-matching, Teacher Sampling Steps=250-step Euler sampling2026.05 | 12.44 |