Efficient and Training-Free Single-Image Diffusion Models

Arxiv CS.CV (computer vision)

Proposes a method to generate images matching a single reference image's patch distribution without any training. Achieves faster generation than prior training-based approaches while maintaining quality.

diffusion-modelstext-to-image

Arxiv CS.CV (computer vision)·Haojun QiuKiriakos N. KutulakosDavid B. Lindell

7 days ago