These are done using the PyTorch Comp-Vis model with the released model weights, the previous dream of tears posts were done using a Keras implementation with quantized model weights on a mac. There is something about the quality and feel of the 2 different generative image synthesis implementations output that i'm still trying to wrap my mind around. The animations in the earlier posts were done using Transition Context processing in Studio Artist off of static generated output images since the mac version is sooooo slow right now. The PyTorch Comp-Vis generative model is running the animations in this post (via processing on google cloud).
No comments:
Post a Comment