IB-Flow: Information Bottleneck-Guided CFG Distillation for Few-Step Text-to-Image Generation

2026-07-10

AI Summary

Existing few-step text-to-image generation methods suffer from over-guidance artifacts and suboptimal performance due to their reliance on static guidance strength and uniform time-step sampling, which overlook the dynamic nature of the generative process. This work addresses this limitation by formulating few-step distillation as a constrained dynamic mutual information optimization problem grounded in the information bottleneck principle. We introduce a dual-track adaptive mechanism that jointly performs instance-aware time-step selection and entropy-aware guidance strength scheduling. To our knowledge, this is the first approach to integrate information-theoretic principles into Classifier-Free Guidance (CFG) distillation, enabling zero-overhead dynamic optimization. Under an extreme two-step setting, our method substantially suppresses artifacts and achieves state-of-the-art generation fidelity, surpassing the current performance ceiling of few-step diffusion compression.

Abstract

While large-scale text-to-image generative models have achieved unprecedented visual performance, their inherent reliance on multi-step iterative solvers incurs severe inference latency. Few-step distillation targeting the Classifier-Free Guidance (CFG) trajectory has emerged as the prevalent dual-dimensional compression paradigm. However, existing frameworks remain subjugated by a coarse-grained blind injection paradigm that perpetually enforces a globally static guidance strength while indiscriminately sampling the supervisor timestep. This state-agnostic design completely disregards the intrinsic nature of image generation as a dynamic evolutionary process characterized by progressive entropy reduction, which not only restricts the performance boundary of few-step compression but also precipitates severe CFG over-conditioning artifacts. To transcend these limitations, we re-examine the distillation procedure through the theoretical lens of Information Theory, formally modeling it as a dynamic mutual information game constrained by the Information Bottleneck (IB) principle. Specifically, we dismantle traditional blind assumptions via a dual-track adaptive framework. To determine the injection target, we propose an instance-aware selection mechanism that transmutes the intractable KL divergence constraint into a zero-overhead closed-form solution predicated on the local vector field norm. To regulate the injection strength, we introduce an entropy-aware schedule that dynamically decays alongside the SNR, applying maximal thrust for initial structural anchoring before smoothly reverting to the natural manifold to refine micro-details. Extensive empirical evaluations corroborate that our framework fundamentally eradicates over-conditioning artifacts, shattering the performance ceiling to achieve SOTA generative fidelity under extremely stringent 2-step configurations.

Problem

• few-step generation
• Classifier-Free Guidance
• distillation
• over-conditioning
• dynamic process

Innovation

• Information Bottleneck
• Classifier-Free Guidance
• Few-Step Distillation
• Adaptive Guidance
• Text-to-Image Generation