Adversarial Learning of Classifier-Free Guidance Schedules

2026-08-14

AI Summary

This study addresses the suboptimality and artifact-prone nature of static classifier-free guidance in diffusion models by proposing a dynamic guidance scale prediction mechanism based on density ratio estimation. Leveraging an adversarial training framework, the method employs a discriminator and a lightweight generative network to jointly learn optimal guidance scales conditioned on time steps, textual prompts, and noisy samples, thereby replacing conventional static global configurations. Experimental results demonstrate that this automated strategy significantly outperforms existing heuristic and dynamic learning approaches on text-to-image generation benchmarks. The proposed approach effectively enhances text-image alignment and generation quality, achieving adaptive optimization of guidance strategies without introducing visual artifacts.

Abstract

Modern text-to-image diffusion models rely on classifier-free guidance (CFG) to achieve high image fidelity and text alignment. However, CFG typically applies a static, global scale across all timesteps, samples, and conditions -- a choice that is generally suboptimal and can introduce artifacts, as different states may benefit from different levels of guidance. While time-varying schedules are known to improve quality, designing them by hand is non-trivial and application-dependent. In this paper, we learn the guidance schedule as a function of diffusion time, conditioning and the current noisy sample, in order to better align sampled images with the text prompt. We frame this as a density ratio estimation problem: a discriminator is trained to estimate the time-dependent log-density ratio between the true and guided marginal distributions, while a lightweight generator network predicts the optimal, state-dependent guidance scale. Empirically, our approach outperforms both heuristic CFG schedules and prior methods for learning dynamic guidance on text-to-image generation benchmarks.

Problem

• Classifier-Free Guidance
• Text-to-Image Diffusion Models
• Dynamic Guidance Schedule
• Density Ratio Estimation

Innovation

• Classifier-Free Guidance
• Adversarial Learning
• Density Ratio Estimation
• Dynamic Guidance Schedule
• Text-to-Image Generation