can vace composite pose and depth video into one video？

[OrangeTreeDev]

hi，can vace composite pose and depth video into one video？ i want use pose and depth processor as source video

[OrangeTreeDev]

if yes，please tell me how to do，thanks lot

[Yuancheng-Xu]

I am wondering the same. Clearly someone has done it (like https://github.com/deepbeepmeep/Wan2GP/issues/347, which seems a little empirical) but there is no official documentation on this.

I wonder what is the correct way to compose depth and pose together.

[rocky533]

There is 2 ways to do this.

1 is using a blend node to blend pose/depth. This can sometimes causes open pose to bleed into the output. 2 is use 2 vace controls. One for each and use only the latent from one of them.

https://openart.ai/workflows/rocky533/wan-21-vace-batched-video-to-video/01Ouu12wBn6XQtH24WqQ

[Yuancheng-Xu]

@rocky533 Thanks for the suggestions! For the second approach, could you elaborate a little bit more? In the Video Condition Unit (VCU) in VACE, it seems to only accept one context video frames. How to use two vace controls?

[rocky533]

if you use same ref image for 2 vace encode nodes, you can connect depth to one, and pose to another. use either of the latent outputs, then pass conditioning from one to the other. If you decode the latents, they are not the control video. They are the ref image and empty latents. The conditioning has the control videos attached and passed through both. This allows you to set strength for each.