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

Source: https://huggingface.co/papers/2606.20404

Abstract

FlowBender is a closed-loop framework that addresses constraint satisfaction in diffusion and flow models by training networks to correct alignment errors using inference-time feedback, outperforming traditional supervised and guidance-based approaches across multiple tasks.

Conditional diffusion andflow modelsroutinely 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 ownalignment error. We introduce FlowBender, aclosed-loop frameworkthat treats this error as a first-class input, training the network to learn a correction policy conditioned on inference-time feedback. At each step, anunguided look-ahead passestimates the clean signal, a task-specific deviation is computed via the forward operator, and arefinement passconsumes this signal to produce a corrected velocity. We propose several variants of FlowBender, including agradient-based formulationfor differentiable operators and azero-order variantfor 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. Acrossimage-to-image translation,restoration, and3D mesh texturing, FlowBender consistently outperforms standard supervised baselines, alignment-loss-augmented training, and state-of-the-artinference-time guidance, improving fidelity and plausibility simultaneously rather than trading them against each other. Project page: https://flow-bender.github.io/

View arXiv pageView PDFProject pageAdd to collection

Get this paper in your agent:

hf papers read 2606\.20404

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.20404 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.20404 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.20404 in a Space README.md to link it from this page.

Collections including this paper1