WanVideo sliding context window test on the 480p 14B T2V -model, worked surprisingly well but incredibly slow (50 mins on a 5090, 513 frames at 832x480)
As an impatient person, I often get obsessed with optimizations, tried implementing TeaCache for WanVideo, failed, but in the process accidentally (maybe) succeeded? Results look promising at least!
WanVideo has been lot of fun, currently playing with applying sliding context windowing similar to AnimateDiff, the 1.3B model is especially suitable for this due to it's speed, here's 1025 frames in one go, under 5GB VRAM used, but took about 10 mins on a 5090 with 30 steps.
HunhuyanVideo -model's vid2vid passes the hippo test! Very promising and versatile model, thanks to the cfg distillation and every possible optimization I could think of this clip of 101 frames at 768x432 took about 2 minutes to sample, fitting slightly under 20GB using my nodes.
