Congrats on the new version of Lightning and the Phased DMD paper! The improved diversity and motion in distillation align with our recent effort on rCM, which employs continuous-time consistency models for better diversity.
I wonder if the full-parameter Wan2.1-T2V-14B checkpoint could be released for comparison. Moreover, it would be great if the diversity evaluation scripts (DINOv3, LPIPS) could be released.
Congrats on the new version of Lightning and the Phased DMD paper! The improved diversity and motion in distillation align with our recent effort on rCM, which employs continuous-time consistency models for better diversity.
I wonder if the full-parameter Wan2.1-T2V-14B checkpoint could be released for comparison. Moreover, it would be great if the diversity evaluation scripts (DINOv3, LPIPS) could be released.