Abstract
Achieving fine-grained control over subject identity and semantic attributes (pose, style, lighting) in text-to-image generation, particularly for multiple subjects, often undermines the editability and coherence of Diffusion Transformers (DiTs). Many approaches introduce artifacts or suffer from attribute entanglement. To overcome these challenges, we propose a novel multi-subject controlled generation model XVerse. By transforming reference images into offsets for token-specific text-stream modulation, XVerse allows for precise and independent control for specific subject without disrupting image latents or features. Consequently, XVerse offers high-fidelity, editable multi-subject image synthesis with robust control over individual subject characteristics and semantic attributes. This advancement significantly improves personalized and complex scene generation capabilities.
Abstract (translated)
在从文本生成图像的过程中，实现对主题身份和语义属性（如姿态、风格、光照）的精细控制，特别是在涉及多个主体时，通常会削弱扩散变压器（DiT）的可编辑性和连贯性。许多方法引入了伪影或遭受属性纠缠的问题。为了解决这些挑战，我们提出了一种新的多主题可控生成模型XVerse。通过将参考图像转换为用于特定标记文本流调节的偏移量，XVerse允许对特定主体进行精确且独立的控制，而不破坏图像潜在特征。因此，XVerse提供了高保真度、可编辑的多主体图像合成，并具有强大的个体主体特性和语义属性控制能力。这一进展显著提高了个性化和复杂场景生成的能力。
URL
https://arxiv.org/abs/2506.21416
