I have a quick question about the library. I'm currently working on a project whose goal is continuously generate video in response to a changing stream of images. Meaning, first a video is generated, then as soon as it's finished, I want to generate another video from the point at which the previous video was generated.
One idea is to try passing in the previous videos through a VAE, noising the output, and passing those embeddings into the "latent" parameter. Is there a better way to do this?