About the loss

[LeningWang]

How does the loss function behave, and after how many steps train can we generate an image instead of noise? Thanks!

[Doubiiu]

Hi, since we build our model based on pre-trained t2v model VideoCrafter. it can already generate videos at the beginning.

[LeningWang]

Thank you! If I use the base architecture of DynamiCrafter to train on our own dataset without using the existing pre-trained parameters, is that feasible? Approximately how many steps would the training take? Thanks again!

[Doubiiu]

Emmm I am not sure. Usually, we will train models on pre-trained weights to make our life easier.