Apr, 2024

InstantStyle:文本到图像生成中面向保留风格的免费午餐

TL;DRTuning-free diffusion-based models have achieved promising results in image personalization and customization, but still face challenges in style-consistent image generation. This paper introduces InstantStyle, a framework that addresses these challenges by decoupling style and content, and injecting reference image features into style-specific blocks to achieve superior visual stylization outcomes.