FramePack is a progressive neural network structure for generating the next frame (partial next frame) prediction in videos.

FramePack compresses input context into a constant length, so that the workload for generation is unaffected by video length.

Even on a laptop GPU, FramePack can handle a large number of frames using a 13B model.

FramePack can be trained with larger batch sizes, similar to batch sizes used in image diffusion training.

Video diffusion, but feels like image diffusion.