New AI Framework Enables 30x Faster High-Quality Image Generation
• Researchers introduced a new framework, distribution matching distillation (DMD), that simplifies the multi-step image generation process of diffusion models down to a single step.
• DMD retains the image quality of more complex diffusion models like Stable Diffusion, while allowing 30 times faster image generation.
• DMD works by training a new network to minimize the difference between its generated images and real images used to train diffusion models.
• In tests, DMD performed on par with state-of-the-art diffusion models in image quality and diversity.
• DMD could enhance design tools by enabling quicker content creation and advancements in areas like drug discovery where speed is important.