Be Yourself: Bounded Attention for Multi-subject Text-to-Image Generation

Keyphrases

• Multiple Subjects 100%
• Text-to-image Generation 100%
• Between-subject 50%
• Information Flow 50%
• Visual Features 50%
• Sampling Process 50%
• Token 50%
• User Control 50%
• Diffusion Model 50%
• Denoising Process 50%
• High Quality Images 50%
• Multi-subject 50%
• Attention Layer 50%
• Training-free 50%
• Text-to-image Diffusion Model 50%
• Layout-to-Image 50%
• Image Extension 50%

Computer Science

• Image Synthesis 100%
• Diffusion Model 100%
• Sampling Process 50%
• Visual Feature 50%
• de-noising 50%
• Image Quality 50%
• Attention (Machine Learning) 50%