New research could make weird AI images a thing of the past

Rice University researchers have developed a new method called ElasticDiffusion to improve AI-generated images. The approach separates local and global image information, allowing for better generation of non-square images at different aspect ratios. This addresses problems with current AI models that often produce visual imperfections when creating non-square images. While ElasticDiffusion currently takes 6-9 times longer to generate images compared to other models, it produces cleaner results without requiring additional training. The researchers aim to reduce processing time to match other popular AI image generators.