cs.CVMar 19, 2026

InjectFlow: Weak Guides Strong via Orthogonal Injection for Flow Matching

Authors: Dayu Wang, Jiaye Yang, Weikang Li, Jiahui Liang, Yang Li

Organizations: Baidu Inc. · Peking University

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

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.

Figures & tables

Figure 1: Qualitative comparison of generation results. While the baseline model often succumbs to trajectory collapse, failing to capture specific attribute bindings or complex spatial relationships, our proposed method effectively rectifies these erroneous trajectories, ensuring accurate semantic alignment with the given prompts.
Figure 2: Empirical observation of latent volume collapse across ODE integration steps.
Figure 3: Our Method
TaskODE ErrorsOurs CorrectedCorrection Rate (%)
Position803442.50
Color & Attr443375.00
Two Objects77100.00
Colors12758.33
Counting221881.82
Table 1: Correction success rate of our method on the GenEval benchmark compared to baseline ODE failures.
Figure 4: Qualitative comparison showing our method successfully correcting attribute binding and color bleeding issues.
Figure 5: Qualitative comparison demonstrating accurate numerical generation for the counting task.
Figure 6: Qualitative comparison highlighting successful corrections of complex spatial relationships.
TaskODE ErrorsOurs CorrectedCorrection Rate (%)
Position742533.78
Color & Attr392564.10
Two Objects99100.00
Colors181266.67
Counting302686.67
Table 2: Correction success rate using Seed 2025.
TaskODE ErrorsOurs CorrectedCorrection Rate (%)
Position773241.56
Color & Attr402870.00
Two Objects55100.00
Colors13753.85
Counting272488.89
Table 3: Correction success rate using Seed 2026.
More in this topic

Flow Matching

Flow matching learns continuous-time vector fields that transport simple noise distributions into data for efficient generative modeling.

See all 656 papers

Explore similar work

CardsList
May 14, 2026

The Velocity Deficit: Initial Energy Injection for Flow Matching

While Flow Matching theoretically guarantees constant-velocity trajectories, we identify a critical breakdown in high-dimensional practice: the Velocity Deficit. We show that the MSE objective systematically underestimates velocity magnitude, causing generated samples to fail to reach the data manifold-a phenomenon we term Integration Lag. To rectify this, we propose Initial Energy Injection, instantiated via two complementary methods: the training-based Magnitude-Aware Flow Matching (MAFM) and the training-free Scale Schedule Corrector (SSC). Both are grounded in our discovery of a crucial asymmetry: velocity contraction causes harmful kinetic stagnation at the trajectory's start, yet acts as a beneficial denoising mechanism at its end. Empirically, SSC yields significant efficiency gains with zero retraining and just one line of code. On ImageNet-1k (256x256), it improves FID by 44.6% (from 13.68 to 7.58) and achieves a 5x speedup, enabling a 50-step generator (FID 7.58) to beat a 250-step baseline (FID 8.65). Furthermore, our methods generalize to Text-to-Image tasks and high-resolution generation, improving FID on MS-COCO by ~22%.

Linze Li, Zong-Wei Hong, Shen Zhang +4

Jiiov Technology, Beijing, China.

Flow MatchingImage GenerationFlow-Based Generative ModelingDiffusion Model Inference Acceleration
AskPDF
Oct 2, 2026

Contextual Flow Matching: Adaptive Step Selection in Flow Models for Efficient Visual Generation

Flow Matching enables high-quality visual generation via continuous-time dynamics, but inference remains costly due to multiple sequential function evaluations. Existing acceleration methods reduce the number of function evaluations but often introduce additional training overhead, degrade quality, or fail to account for input-dependent variability. We propose COFLOW, an inference-time method that adaptively selects the step counts each generation based on the prompt features. Our context-aware COFLOW is trained online with an unsupervised reward that balances inference efficiency and generation fidelity. Our method is plug-and-play, requiring no retraining of the underlying generative model. It generalizes to image and video generation, achieving over 2.5x speedup while preserving perceptual and semantic quality. We further provide a theoretical analysis establishing an O(1/K) forward-Euler discretization error bound under standard regularity conditions.

Divya Jyoti Bajpai, Arun Verma, Manjesh Kumar Hanawal

Department of IEOR IIT Bombay, Mumbai, India · SMART Centre Singapore

Efficient InferenceFlow MatchingImage GenerationFlow-Based Generative Modeling
AskPDF
Jun 9, 2026

Mean Flow Distillation: Robust and Stable Distillation for Flow Matching Models

Flow Matching models have demonstrated strong performance across a wide range of generative tasks. However, their reliance on ODE-based iterative sampling incurs substantial computational overhead in inference, which limits their applicability in real-time scenes. While distillation is a promising solution, existing approaches largely borrow from diffusion-based score matching, often failing to exploit the intrinsic geometric structure of flows and suffering from training instability, high variance, and degraded generation quality. In this paper, we propose Mean Flow Distillation (MFD), a novel distillation framework tailored for flow matching models. We theoretically demonstrate that MFD acts as a temporal low-pass filter, effectively suppressing the high-frequency optimization noise inherent in variational score distillation (VSD) while ensuring global trajectory consistency. We further prove the Mean Flow Matching Theorem, establishing that matching expected average velocities is sufficient for strict distribution alignment. Empirically, on challenging tasks of high-dimensional manifolds including 4D occupancy forecasting and text-to-image generation, MFD achieves state-of-the-art performance, enabling high-fidelity single-step generation.

An Zhao, Shengyuan Zhang, Zhongjian Sun +5

College of Computer Science and Technology, Zhejiang University, Zhejiang, China · School of Software Technology, Zhejiang University, Zhejiang, China · Princeton University, Princeton, USA +1

Flow MatchingOne-Step Generative ModelingFlow-Based Generative Modeling
AskPDF