FlowConsist: Make Your Flow Consistent with Real Trajectory

2026-02-06

AI Summary

Existing fast flow models suffer from trajectory drift and error accumulation due to their reliance on randomly paired noise–data samples to construct conditional velocities, causing significant deviation from the true ODE trajectories. To address this, this work proposes FlowConsist, a novel framework that explicitly optimizes for trajectory consistency. FlowConsist abandons inconsistent conditional velocity constructions and instead leverages marginal velocities predicted by the model itself, augmented with a time-step-wise distribution alignment mechanism to correct trajectories during generation. This ensures that the sampling process evolves along a consistent path aligned with the underlying ODE dynamics. The method achieves a state-of-the-art single-step generation performance, attaining an FID of 1.52 on ImageNet 256×256 with only one sampling step.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Consent frameworks and practices on the web

Abstract

Fast flow models accelerate the iterative sampling process by learning to directly predict ODE path integrals, enabling one-step or few-step generation. However, we argue that current fast-flow training paradigms suffer from two fundamental issues. First, conditional velocities constructed from randomly paired noise-data samples introduce systematic trajectory drift, preventing models from following a consistent ODE path. Second, the model's approximation errors accumulate over time steps, leading to severe deviations across long time intervals. To address these issues, we propose FlowConsist, a training framework designed to enforce trajectory consistency in fast flows. We propose a principled alternative that replaces conditional velocities with the marginal velocities predicted by the model itself, aligning optimization with the true trajectory. To further address error accumulation over time steps, we introduce a trajectory rectification strategy that aligns the marginal distributions of generated and real samples at every time step along the trajectory. Our method establishes a new state-of-the-art on ImageNet 256$\times$256, achieving an FID of 1.52 with only 1 sampling step.

Problem

• fast flow models
• trajectory drift
• error accumulation
• ODE path consistency
• marginal velocity

Innovation

• FlowConsist
• trajectory consistency
• fast flow models
• marginal velocity
• trajectory rectification