This workflow is a I2V, it implements Color Match, SageAttention, and is dedicated to make seamless 10s videos on Wan 2.2 model. It is important to keep seed, samplers and schedulers the same according through the whole workflow, also one could make positive prompt empty for better consistency.

This workflow also implements MMaudio. It works some times on RTX 4070 TI Super, but other times it loads audio model simultaneously with WAN, so in goes into memory error. One could turn 2 two last switched to OFF to keep out of this error, or could have a video card with a bigger VRAM.

IF YOU KNOW HOW TO MAKE THE LOADER OF MMAUDIO start after the video is completed, or how we could go with PurgeVram or PurgeVram V2 nodes in any way, please tell me how.

If one has questions, I'd answer gladly.

In previews the only second one goes with audio, and it was made on completely other workflow by Coyotte's Movie maker that is not mine, before the total rework to 5s+5s video plus mmaudio. This workflow is inside the video file if someone wants it.

By the way, MMaudio works in this way: it downscales our video to 384px (or less, i don't remember) by long side, then in crops a square of about 256x256px (or less, I don't remember either), then it looks on it an makes audio sync. The idea is that the moving/sounding part should be in the center of you video. It would surely approximate, it does this bad.

If one doesn't know how to implement sageattention, please comment, it gives for about 12-14% speed improvement on my 4070 ti Super.