HyperFluid is trained on 16 Stable Diffusion generated images. Keywords that work well together are 'turbulence' 'vorticity' 'smoke'. Works well together with PhotoHelper, CinemaHelper, ActionHelper, and HeartArt.
16 images, 4 vectors, batch size of 4, 4 steps gradient accumulation, 300 steps/epochs. Learning rate of 0.003.