Huge Improvement
I was testing WAN 2.2 t2v on my 3090, and it was taking 3 hours to create 121 frames at 720p. Using your models, it took 19 minutes! Thank you very much.
Thank you, greatly appreciated!
That said, I tried it on my 5080 and got the "Allocation on device. This error means you ran out of memory on your GPU." error.
I'm guessing that's because the models are 20 GB, and they can't fit inside the 5080's 16 GB of VRAM.
You must be doing something wrong... I'm running this with 12GB VRAM with no OOM errors. The models include WAN VAE and umt5 CLIP, so the file sizes are misleading. Try using the workflow examples I provided.
I also get OOM on a 4070 Super with 12GB VRAM both in normal comfyui mode and lowmemory comfyui mode. My RAM is only 24GB though. I think for this to work on Low VRAM Cards, we need at least 30 or 40+ GB system ram so it can be offloaded properly by ComfyUI.
You might be able to increase your operating system's swap/paging file size if you are running low on RAM to make things work... maybe.
You are blocking your KSampler settings... are you using a CFG of 1? You are likely running out of memory and swapping hard at 720p and 121 frames.