AI绘画系统有两个重要模块

Contrastive Language Image Pre-training (CLIP)

Diffusion Model

Diffusion - Latent diffusion models (LDMs)

Dec/2021: High-Resolution Image Synthesis with Latent Diffusion Models

Ref: https://arxiv.org/pdf/2112.10752.pdf

Abstract


2021/2022 state-of-art 结果,在当时主要针对pixel space

问题:如果运算资源不够强大,不能产生像样的结果,消费级GPU负担过大

New point:cross-attention layer