When training, I have a few videos, all 16 frames long. When I train on multiple videos there is a slowdown compared to training on one video.
From what I could gather, the expected result would be that it switches between the videos, so the final step count would be the same, and the steps would be spread out over the videos.
But looking at the code (at least at a glance), it would appear that it would be cycling through the images (as intended). But when actually running it, there's a massive slowdown. (Going from 6s/it to 50s/it at a much larger video count).
There appears to have been a mistake, my apologies. It seems to be working now. I'm not sure what was causing it earlier, but it isn't doing it anymore, so it appears to have just been a user issue.
When training, I have a few videos, all 16 frames long. When I train on multiple videos there is a slowdown compared to training on one video.
From what I could gather, the expected result would be that it switches between the videos, so the final step count would be the same, and the steps would be spread out over the videos.
In reality, when training, there is a slowdown the more videos you train on. It also claims that this number is the batch size while from https://github.com/kijai/ComfyUI-ADMotionDirector/issues/6#issuecomment-2012396372 and the fact that itertools is used it appears it is intended to be the batch count, not size.
But looking at the code (at least at a glance), it would appear that it would be cycling through the images (as intended). But when actually running it, there's a massive slowdown. (Going from 6s/it to 50s/it at a much larger video count).