FlowBender: Feedback-Aware Training for Self-Correcting Conditional Flows

2026-06-18

AI Summary

Conditional generative models often struggle to simultaneously achieve high fidelity and plausible generation under task-specific constraints, such as depth map consistency. To address this challenge, this work proposes FlowBender—a closed-loop training framework that, for the first time, treats alignment errors observed during inference as first-class training signals, enabling the model to learn feedback-driven self-correction strategies. FlowBender is compatible with both differentiable and non-differentiable forward operators and introduces a low-overhead prior-step shortcut mechanism. By integrating conditional flow models, bias estimation, and gradient-based or zeroth-order correction strategies, it enables efficient sampling optimization. Experiments demonstrate that FlowBender significantly outperforms supervised baselines, alignment-loss-trained models, and existing guidance methods across image translation, restoration, and 3D texture generation tasks, consistently improving both conditional fidelity and overall generation quality.

Abstract

Conditional diffusion and flow models routinely fail to satisfy the very constraints that define their task. For instance, a depth-conditioned model often produces images whose re-extracted depth disagrees with the input, even though the forward operator--the depth predictor defining the constraint--is available during both training and inference. Existing approaches generally fall into two categories: supervised models that treat the conditioning signal as a static cue and ignore alignment information at inference, and guidance-based methods that consult it through hand-tuned linear updates, typically trading fidelity to the condition against the plausibility of the generated sample. We argue that the fundamental gap in both paradigms is that the model is never trained to utilize its own alignment error. We introduce FlowBender, a closed-loop framework that treats this error as a first-class input, training the network to learn a correction policy conditioned on inference-time feedback. At each step, an unguided look-ahead pass estimates the clean signal, a task-specific deviation is computed via the forward operator, and a refinement pass consumes this signal to produce a corrected velocity. We propose several variants of FlowBender, including a gradient-based formulation for differentiable operators and a zero-order variant for non-differentiable settings such as JPEG compression. For efficient sampling, we introduce a prior-step shortcut that enables closed-loop correction at a minimal additional computational cost. Across image-to-image translation, restoration, and 3D mesh texturing, FlowBender consistently outperforms standard supervised baselines, alignment-loss-augmented training, and state-of-the-art inference-time guidance, improving fidelity and plausibility simultaneously rather than trading them against each other. Project page: https://flow-bender.github.io/

Problem

• conditional flows
• constraint satisfaction
• alignment error
• feedback-aware training
• self-correction

Innovation

• self-correcting flows
• feedback-aware training
• conditional generation
• closed-loop correction
• alignment error