You can use FreeNoise noise_type to increase consistency between context windows
Hey, just wanted to mention that when using AnimateDiff-Evolved, you can attach the Sample Settings node and set noise_type to FreeNoise to make the context changes less drastic! All FreeNoise does is basically reuse the same initial noise in the context overlaps as the first context_length window, and places a randomly scrambled version in each overlap of context windows from that point onwards. The repeated_context noise_type does the same, but does not scramble the noise in the context window overlaps, which has the side effect of making the contents repeat every context_length frames.
Amazing. Let me try it right now.
This is way better. I updated the workflow json. Thank you!
@Kosinkadink Btw, is there a way for the video loader to report the fps? So that we can save the video using the same fps.
@PeterL1n
not yet, but that is an enhancement AustinMroz and I will soon add - there is nothing on a technical level that prevents it from being implemented, but there was just a lot of higher-priority tasks that kept taking up time. The goal would be to have a video_info output that could connect to a Video Info node, which would then break up into a bunch of outputs for both the original video details, and the 'actual' loaded video details like fps, frame count, etc.