[contingo]

SALT_VERSE just uses still images generated by DALL•E-2 and Midjourney. They're animated using techniques that have already been around for a long time: panning around and zooming in and out of a larger frame, fake raster-to-3d effects (auto projection mapping), typical After Effects transitions. There are some overlaid face animations done with something along the lines of NVIDIA's Audio2Face. None of it is video generated by SD, DALL•E-2 and Midjourney.

There have been some actual video explorations of SD's latent space, this example amazes me: https://twitter.com/karpathy/status/1559343616270557184