Search papers, labs, and topics across Lattice.
This paper investigates the effectiveness of training-free guidance (TFG) in pretrained diffusion models and introduces a novel approach using $x$-prediction, which directly outputs clean images, thereby minimizing errors associated with estimating clean images from noisy states. The authors provide a theoretical analysis of how different prediction targets affect accuracy and introduce a new metric, guided-class FID (Child FID), to evaluate the impact of guidance on manifold integrity. Experimental results demonstrate that $x$-prediction significantly outperforms traditional methods in maintaining sample fidelity on the manifold, particularly in challenging tasks like fine-grained bird recognition and style transfer.
$x$-prediction outperforms traditional methods by directly generating clean images, ensuring that guided samples remain on the data manifold even in high-noise conditions.
Training-free guidance (TFG) steers a pretrained diffusion model toward a desired attribute at inference. To be effective, this guidance must be applied from the earliest, high-noise steps of sampling. Because its objective (a classifier or energy) is defined on clean images, $蔚$- and $v$-prediction models must first estimate the clean image $\hat{x}$ from the noisy state at each step, and the accuracy of that estimate determines how easily guidance drifts off the data manifold. $x$-prediction, a recent alternative, outputs the clean image directly, removing this source of error even at high noise. This is our motivation. We provide a theoretical analysis of how each prediction target shapes this accuracy, and introduce guided-class FID (Child FID), a metric that exposes the manifold damage standard evaluation misses. Experiments on a new fine-grained bird benchmark and on style transfer confirm that $x$-prediction keeps guided samples on the manifold most reliably, making it the strongest foundation for training-free guidance. Code is available at https://github.com/ManLuML/on-manifold-tfg
