Probably a pose model. I don’t know what they do beyond that but the pi ai camera has a pose, and segmentation model it comes with in the demo software and it can do that shit real time with practically 0 hardware use on the pi.
I messed around in comfy ui with the pose stuff before but only with static images, so i imagine video is a similar workflow you just need more vram to use those models.
How they do it? It seems that they first map the body movement with something like blender, then they just replace the virtual character “skin”
AI.
They do none of that
Probably a pose model. I don’t know what they do beyond that but the pi ai camera has a pose, and segmentation model it comes with in the demo software and it can do that shit real time with practically 0 hardware use on the pi.
I messed around in comfy ui with the pose stuff before but only with static images, so i imagine video is a similar workflow you just need more vram to use those models.