InjectFlow: Weak Guides Strong via Orthogonal Injection for Flow Matching

TL;DR

InjectFlow enhances flow matching models by injecting orthogonal semantics to mitigate dataset bias effects, significantly improving out-of-distribution sample generation without retraining.

Contribution

The paper introduces InjectFlow, a training-free method that injects orthogonal semantics to address bias in flow matching models, backed by theoretical analysis and extensive experiments.

Findings

Fixes 75% of failed prompts on GenEval dataset

Prevents trajectory lock-in caused by bias manifold

Maintains high generative quality while improving fairness

Abstract

Flow Matching (FM) has recently emerged as a leading approach for high-fidelity visual generation, offering a robust continuous-time alternative to ordinary differential equation (ODE) based models. However, despite their success, FM models are highly sensitive to dataset biases, which cause severe semantic degradation when generating out-of-distribution or minority-class samples. In this paper, we provide a rigorous mathematical formalization of the ``Bias Manifold'' within the FM framework. We identify that this performance drop is driven by conditional expectation smoothing, a mechanism that inevitably leads to trajectory lock-in during inference. To resolve this, we introduce InjectFlow, a novel, training-free method by injecting orthogonal semantics during the initial velocity field computation, without requiring any changes to the random seeds. This design effectively prevents the latent drift toward majority modes while maintaining high generative quality. Extensive experiments demonstrate the effectiveness of our approach. Notably, on the GenEval dataset, InjectFlow successfully fixes 75% of the prompts that standard flow matching models fail to generate correctly. Ultimately, our theoretical analysis and algorithm provide a ready-to-use solution for building more fair and robust visual foundation models.