The Rundown: Researchers at MIT just introduced a new method called ‘Distribution Matching Distillation’ — enabling faster AI image generation while maintaining the quality of the original model.
The details:
DMD distills the knowledge of a multi-step model (like Stable Diffusion) into a faster model that generates images in a single step.The method utilizes two diffusion models as guides to train the new, fast model, ensuring stability and consistency. In tests, DMD generated images 30 times faster than Stable Diffusion while maintaining or surpassing the quality level of outputs.
Why it matters: By dramatically reducing the computational cost and time required for high-quality image generation, DMD could unlock new possibilities for tools like real-time visual editing, design tools, drug discovery, and 3D modeling applications.
See the full story here: https://news.mit.edu/2024/ai-generates-high-quality-images-30-times-faster-single-step-0321?utm_source=www.therundown.ai&utm_medium=newsletter&utm_campaign=apple-s-ai-master-plan