Search papers, labs, and topics across Lattice.
This paper introduces Self-Swap Guidance (SSG), a novel technique to enable classifier-free guidance (CFG)-like control in both conditional and unconditional diffusion model generation. SSG perturbs predictions by swapping semantically dissimilar token latents in either spatial or channel dimensions, using the difference between perturbed and clean predictions to guide sampling. Experiments on MS-COCO and ImageNet demonstrate that SSG outperforms existing condition-free methods in image fidelity and prompt alignment, while also improving robustness due to its fine-grained perturbation.
Unlock CFG-like guidance for unconditional diffusion models with Self-Swap Guidance (SSG), a surprisingly simple token-swapping technique that boosts image quality and prompt alignment.
Classifier-Free Guidance (CFG) is a widely used inference-time technique to boost the image quality of diffusion models. Yet, its reliance on text conditions prevents its use in unconditional generation. We propose a simple method to enable CFG-like guidance for both conditional and unconditional generation. The key idea is to generate a perturbed prediction via simple token swap operations, and use the direction between it and the clean prediction to steer sampling towards higher-fidelity distributions. In practice, we swap pairs of most semantically dissimilar token latents in either spatial or channel dimensions. Unlike existing methods that apply perturbation in a global or less constrained manner, our approach selectively exchanges and recomposes token latents, allowing finer control over perturbation and its influence on generated samples. Experiments on MS-COCO 2014, MS-COCO 2017, and ImageNet datasets demonstrate that the proposed Self-Swap Guidance (SSG), when applied to popular diffusion models, outperforms previous condition-free methods in image fidelity and prompt alignment under different set-ups. Its fine-grained perturbation granularity also improves robustness, reducing side-effects across a wider range of perturbation strengths. Overall, SSG extends CFG to a broader scope of applications including both conditional and unconditional generation, and can be readily inserted into any diffusion model as a plug-in to gain immediate improvements.
