DisEnvisioner: Disentangled and Enriched Visual Prompt for Customized Image Generation

ABSTRACT In the realm of image generation, creating customized images from visual prompt with additional textual instruction emerges as a promising endeavor. However, existing methods, both tuning-based and tuning-free, struggle with interpreting the subject-essential attributes from the visual prompt. This leads to subject-irrelevant attributes infiltrating the generation process, ultimately compromising the personalization quality in both editability and ID preservation. In this paper, we present DisEnvisioner, a novel approach for effectively extracting and enriching the subject-essential features while filtering out -irrelevant information, enabling exceptional customization performance, in a tuning-free manner and using only a single image. Specifically, the feature of the subject and other irrelevant components are effectively separated into distinctive visual tokens, enabling a much more accurate customization. Aiming to further improving the ID consistency, we enrich the disentangled features, sculpting them into a more granular representation. Experiments demonstrate the superiority of our approach over existing methods in instruction response (editability), ID consistency, inference speed, and the overall image quality, highlighting the effectiveness and efficiency of DisEnvisioner. (Ruiz et al., 2023; Chen et al., 2023a; Li et al., 2024; Wei et al., 2023; Ye et al., 2023) under single-image setting. We evaluate these methods on the same subject with different poses and environments. It can be observed that irrelevant factors, such as the subject's posture ( ) and background ( ), can affect the customization quality and result in poor editability ( ) or poor ID consistency ( ). For instance, BLIP-Diffusion falls in both two factors, leading to poor editability and ID consistency. We denotes its performance as " " (the symbols without color filling, such as " ", indicates that the customization is not affected by subject's posture, and " " indicates good editability). We also try to use masks to filter out irrelevant information for these methods. However, the harmful influence of subject's posture still exists. And solid background colors (e.g., white or black) also can harmfully impact the customization quality, leading to textureless backgrounds.