Learning Adapters for Text-Guided Portrait Stylization with Pretrained Diffusion Models

This paper presents a framework for text-guided face portrait stylization using a pre-trained large-scale diffusion model. To balance style transformation and content preservation, we introduce an adapter that modifies specific components of the diffusion model. By training the adapter to only modify these components, we reduce the tuning parameter space, resulting in an efficient solution for face portrait stylization. Our approach captures the target style and at the same time, preserves the source portrait content, making it an effective method for personalized image editing. Experimental results show its superiority over state-of-the-art techniques in various stylization tasks.