Generative Modeling via Drifting

2026-02-04

AI Summary

This work proposes Drifting Models, a novel generative paradigm that overcomes the reliance on multi-step iterative inference inherent in diffusion and flow models by enabling high-quality single-step generation. The approach dynamically evolves the pushforward distribution during training and employs a neural network optimizer-driven drift field to guide sample trajectories, naturally aligning the generated distribution with the target distribution. Its key innovation lies in embedding the distribution evolution mechanism directly into the training process, thereby endowing the model with native single-step generation capability. Experimental results demonstrate state-of-the-art performance, achieving a latent-space FID of 1.54 and a pixel-space FID of 1.61 on ImageNet at 256×256 resolution.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Generative modeling can be formulated as learning a mapping f such that its pushforward distribution matches the data distribution. The pushforward behavior can be carried out iteratively at inference time, for example in diffusion and flow-based models. In this paper, we propose a new paradigm called Drifting Models, which evolve the pushforward distribution during training and naturally admit one-step inference. We introduce a drifting field that governs the sample movement and achieves equilibrium when the distributions match. This leads to a training objective that allows the neural network optimizer to evolve the distribution. In experiments, our one-step generator achieves state-of-the-art results on ImageNet at 256 x 256 resolution, with an FID of 1.54 in latent space and 1.61 in pixel space. We hope that our work opens up new opportunities for high-quality one-step generation.

Problem

• generative modeling
• one-step inference
• pushforward distribution
• data distribution matching
• high-quality generation

Innovation

• Drifting Models
• one-step inference
• pushforward distribution
• generative modeling
• drifting field