Last released Oct 24, 2026
A 256px text-to-image model for face generation: single-stream rectified-flow DiT over an f16x96 VAE latent.