Researchers from UCSD and Adobe Introduce Presto!: An AI Approach to Inference Acceleration for Score-based Diffusion Transformers via Reducing both Sampling Steps and Cost Per Step