Implementation of the 'trajectory_imitation'

[taey16]

Dear DiffSynth-Studio team

First of all, thanks a lot for your kindly support in advance.

I reviewed your implementation about 'trajectory_imitation'.
Did you have any successful story (or demonstration) with the 'trajectory_imitation'??

In my idea, the algorithm produces multiple DMD(Distribution Matching Distillation)-gradients across a whole trajectory in a training-step, which would be result in a competitive performance comparing to the plain-DMD. But, I didn't find any explanation of your implementation or paper, etc.

Plz, Let me know anything about your implementation of 'trajectory_imitation'.

Best,

From a github user.