FramePack can process a very large number of frames with 13B models even on laptop GPUs. FramePack can be trained with a much larger batch size, similar to the batch size for image diffusion training.