What architecture enables pose-consistent, photorealistic virtual outfit changes?

How about VTONs ?