Improving image quality and variation in diffusion models without compromising alignment with given conditions, such as class labels or text prompts, is a significant challenge. Current methods often enhance image quality at the expense of diversity, limiting their applicability in various real-world scenarios such as medical diagnosis and autonomous driving, where both high quality and…
