Why it matters
FLUX is leading open image model. Understanding shapes current state.
Advertisement
The architecture
Rectified flow: smoother path than diffusion.
MMDiT: multimodal DiT combines text + image tokens.
Advertisement
How it works end to end
Rectified flow: fewer sampling steps.
MMDiT: joint text + image attention.
Variants: dev, schnell (distilled), pro (API).