[chacham15]

I dont think "turn the horse's head 90˚" is the right path forward. What I think is more likely and more useful is: here is a start keyframe and here is a stop keyframe (generated by text to image using other things like controlnet to control positioning etc.) and then having the AI generate the frames in between. Dont like the way it generated the in between? Choose a keyframe, adjust it, and rerun with the segment before and segment after.

[GenerocUsername]

This appeals to me because it feels auditable and controllable... But the pace these things have been progressing the last 3 years, I could imagine the tech leapfrogs all conventional understanding real soon. Likely outputting gaussian splat style outputs where the scene is separate from the camera and ask peices can be independently tweaked via a VR director chair

[8note]

So a declarative keyframe of "the horses head is pointed forward" and a second one of "the horse is looking left"

And let the robot tween?

Vs an imperative for "tween this by turning the horse's head left"