A technique that performs diffusion in a compressed 'latent space' rather than on pixels, making high-resolution generation computationally feasible.
The architecture behind Stable Diffusion (Rombach et al.).
Democratized AI art generation on consumer GPUs.