One-step Latent-free Image Generation with Pixel Mean Flows

Modern diffusion/flow-based models for image generation typically exhibit two core characteristics: (i) using multi-step sampling, and (ii) operating in a latent space. Recent advances have made encouraging progress on each aspect individually, paving the way toward one-step diffusion/flow without latents. In this work, we take a further step towards this goal and propose "pixel MeanFlow" (pMF). Our core guideline is to formulate the network output space and the loss space separately. The network target is designed to be on a presumed low-dimensional image manifold (i.e., x-prediction), while the loss is defined via MeanFlow in the velocity space. We introduce a simple transformation between the image manifold and the average velocity field. In experiments, pMF achieves strong results for one-step latent-free generation on ImageNet at 256$\times$256 resolution (2.22 FID) and 512$\times$512 resolution (2.48 FID), filling a key missing piece in this regime. We hope that our study will further advance the boundaries of diffusion/flow-based generative models.

Modern AI image generators usually take many steps to produce an image, and they work on a compressed latent representation rather than directly on pixels. We show it is possible to generate high-quality images in a single step, directly in pixel space. Our method combines two recent ideas — predicting an entire trajectory at once, and predicting the clean image rather than the noise direction — into a single training framework. On standard ImageNet benchmarks, our one-step model is competitive with much slower multi-step methods. This suggests the multi-step + latent recipe that has dominated image generation may not be necessary, opening a path toward simpler end-to-end generative models.