Qiu Weimin

Diversifying Similar Subjects for Text-to-image Synthesis with Self-Cross Diffusion Guidance and Reward

IJCV, 2026.

• Diversity is an unresolved issue for diffusion-based image synthesis, particularly for synthesizing multiple subjects of the same category. We identify multiple objects of the same category during generation and propose an instance segmentation method that leverages the attention maps of diffusion models, formulating our guidance between instances. We also adapt our guidance as an effective reward for RL-based post-training and show improved subject diversity with no computational overhead at inference time.

Self-Cross Diffusion Guidance for Image Synthesis of Similar Subjects

CVPR, 2025.

• Link to Publication
• Subject mixing is an unresolved issue for diffusion-based image synthesis, particularly for synthesizing multiple similar-looking subjects. We propose Self-Cross Diffusion Guidance to penalize the overlap between cross-attention maps and the aggregated self-attention map. Compared to previous methods based on self-attention or cross-attention alone, our guidance is training-free and more effective in eliminating subject mixing.