CompVis Releases iDDM for Few-Step Diffusion in Single-Stage Training

CompVis introduced Improved Distributional Diffusion Models (iDDM), which trains from scratch in a single stage without teacher models or self-distillation, achieving a 4-step FID of 4.48 on ImageNet-256.

2026-09-30 ~ 2026-10-02 · 2 related posts

• CompVis improves Distributional Diffusion Models: 4.48 FID at 4 steps on ImageNet — CompVis · 2026-09-30
• iDDM trains few-step diffusion from scratch in one stage: 4.48 FID at 4 steps on ImageNet-256 — proceduralia · 2026-10-02