Stable Diffusion 3.5 vs. FLUX.1: Architectural Deep Dive into MMDiT vs. UNet

For four years, the latent diffusion world was dominated by the classic U-Net architecture. In 2026, the transition to Diffusion Transformers (DiT) is complete. We break down the structural differences between Stability AI's SD 3.5 Large and Black Forest Labs' FLUX.1.

2. How MMDiT Dual-Stream Attention Works

MMDiT maintains two distinct sets of weights for text and image modalities. During attention blocks, representation vectors join into a single sequence, allowing image latents to directly reshape text token weights and vice versa.