Agreed! I wonder if we could use a little fine tuning to enforce more visual consistency between the key frame generations — combined with depth conditioning it seems to take us most of the way there

→ View original post on X — @bilawalsidhu