Browse Teach Curate

Multimodal Diffusion Transformers with Separate Text and Image Streams

the dominant 2026 backbone — dual-stream blocks with separate weights per modality, which is why current models finally render text inside images; volatile, tied to a specific generation of models

jeremy Video

Stable Diffusion 3 MMDiT Rectified Flow Transformers for High-Resolution Image Synthesis

Stable Diffusion 3 employs Multilingual Multimodal DiT (MMDiT) architecture to perform high-resolution image synthesis by defining efficient straight-line flow trajectories between noise and data dis…