In-situ Autoguidance: Eliciting Self-Correction in Diffusion Models

TL;DR

This paper introduces In-situ Autoguidance, a novel method enabling diffusion models to self-correct during inference without auxiliary models, improving image quality and diversity efficiently.

Contribution

The paper proposes a zero-cost, self-guidance technique for diffusion models that eliminates the need for auxiliary models, enhancing efficiency and effectiveness.

Findings

Self-guidance improves image quality and diversity.

The method matches or surpasses auxiliary-guided approaches.

It establishes a new baseline for cost-efficient diffusion guidance.

Abstract

The generation of high-quality, diverse, and prompt-aligned images is a central goal in image-generating diffusion models. The popular classifier-free guidance (CFG) approach improves quality and alignment at the cost of reduced variation, creating an inherent entanglement of these effects. Recent work has successfully disentangled these properties by guiding a model with a separately trained, inferior counterpart; however, this solution introduces the considerable overhead of requiring an auxiliary model. We challenge this prerequisite by introducing In-situ Autoguidance, a method that elicits guidance from the model itself without any auxiliary components. Our approach dynamically generates an inferior prediction on the fly using a stochastic forward pass, reframing guidance as a form of inference-time self-correction. We demonstrate that this zero-cost approach is not only viable but also establishes a powerful new baseline for cost-efficient guidance, proving that the benefits of self-guidance can be achieved without external models.