Angle Domain Guidance: Latent Diffusion Requires Rotation Rather Than Extrapolation
This addresses a specific problem of color distortion in high-quality image synthesis for users of text-to-image models, representing an incremental improvement over existing guidance techniques.
The paper tackled color distortions in text-to-image latent diffusion models under high classifier-free guidance weights by identifying norm amplification as the cause and proposing an Angle Domain Guidance algorithm. The result was significantly improved color fidelity and human perceptual alignment while preserving text-image alignment, as demonstrated experimentally.
Classifier-free guidance (CFG) has emerged as a pivotal advancement in text-to-image latent diffusion models, establishing itself as a cornerstone technique for achieving high-quality image synthesis. However, under high guidance weights, where text-image alignment is significantly enhanced, CFG also leads to pronounced color distortions in the generated images. We identify that these distortions stem from the amplification of sample norms in the latent space. We present a theoretical framework that elucidates the mechanisms of norm amplification and anomalous diffusion phenomena induced by classifier-free guidance. Leveraging our theoretical insights and the latent space structure, we propose an Angle Domain Guidance (ADG) algorithm. ADG constrains magnitude variations while optimizing angular alignment, thereby mitigating color distortions while preserving the enhanced text-image alignment achieved at higher guidance weights. Experimental results demonstrate that ADG significantly outperforms existing methods, generating images that not only maintain superior text alignment but also exhibit improved color fidelity and better alignment with human perceptual preferences.