Three Creates All: You Only Sample 3 Steps

1. 1Freeze the pretrained diffusion backbone
2. 2Distill a small set of step-wise, layer-wise time embeddings from reference data
3. 3Optimize the time embeddings for few-step sampling using MTEO
4. 4Evaluate the performance of the optimized model on generation tasks

AI engineers and researchers can benefit from this technique to improve the efficiency of diffusion models, while ML researchers can apply this to develop more efficient generative models

💡 Standard timestep conditioning is a key bottleneck for few-step sampling in diffusion models, and MTEO can help overcome this limitation

💡 Diffusion models can be optimized for faster inference by sampling only 3 steps!