Core Concepts
NoiseCollage proposes a novel approach to generate multi-object images accurately by independently estimating noises for individual objects and merging them into a single noise.
Abstract
NoiseCollage introduces a unique layout-aware text-to-image diffusion model that addresses issues in existing models by employing a crop-and-merge operation of noises. This innovative approach results in high-quality, accurate image generation with improved layout control. The integration of ControlNet further enhances the model's flexibility and accuracy in generating images with additional conditions like edges, sketches, and pose skeletons. Experimental results demonstrate NoiseCollage's superiority over state-of-the-art methods in layout-aware image generation.
Stats
"Qualitative and quantitative evaluations show that NoiseCollage outperforms several state-of-the-art models."
"Experimental results indicate that the crop-and-merge operation of noises is a reasonable strategy to control image generation."
"The Training-free nature of NoiseCollage allows direct integration with ControlNet and realizes finer output controls by edge images, sketches, and body skeletons."