Playing around with shift attention, I see some potential.
It can't really wok with motion but for emotions and expressions I can work with it.
There might be a way to chain the last one's seed to do the next 'act' but there needs to be some prompt design to make sure to use the right tags to be shifted and lock in which tags become global.
Seems I have a new toy to play with
Low res stuff to see what it can do.
https://files.catbox.moe/21wa0u.mp4https://files.catbox.moe/p8724x.mp4https://files.catbox.moe/lq8gc5.mp4 