Decoupled MeanFlow: Turning Flow Models into Flow Maps for Accelerated Sampling
This work addresses the computational inefficiency of generative models for practitioners, offering a significant speed-up with minimal quality loss, though it is incremental in improving existing flow-based methods.
The paper tackles the problem of slow sampling in denoising generative models like flow models by introducing Decoupled MeanFlow, a method that converts these models into flow maps without architectural changes, enabling high-quality image generation in as few as 1 to 4 steps with FID scores as low as 1.51 on ImageNet 512x512.
Denoising generative models, such as diffusion and flow-based models, produce high-quality samples but require many denoising steps due to discretization error. Flow maps, which estimate the average velocity between timesteps, mitigate this error and enable faster sampling. However, their training typically demands architectural changes that limit compatibility with pretrained flow models. We introduce Decoupled MeanFlow, a simple decoding strategy that converts flow models into flow map models without architectural modifications. Our method conditions the final blocks of diffusion transformers on the subsequent timestep, allowing pretrained flow models to be directly repurposed as flow maps. Combined with enhanced training techniques, this design enables high-quality generation in as few as 1 to 4 steps. Notably, we find that training flow models and subsequently converting them is more efficient and effective than training flow maps from scratch. On ImageNet 256x256 and 512x512, our models attain 1-step FID of 2.16 and 2.12, respectively, surpassing prior art by a large margin. Furthermore, we achieve FID of 1.51 and 1.68 when increasing the steps to 4, which nearly matches the performance of flow models while delivering over 100x faster inference.