Classifier-free Guidance with Adaptive Scaling

Dawid Malarz, Artur Kasymov, Maciej Zieba, Jacek Tabor, Przemysław Spurek · 2025

Read Classifier-free Guidance with Adaptive Scaling by Dawid Malarz, Artur Kasymov, Maciej Zieba, Jacek Tabor, Przemysław Spurek (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-10574.txt.

Description

Dawid Malarz and co-authors refine in 2025 the guidance trick that steers text-to-image diffusion models. Classifier-free guidance trades prompt fidelity against image quality: strong guidance hugs the text but degrades the picture, weak guidance does the reverse. Their variant adapts the guidance strength as generation proceeds.

First, gradient-based normalization steadies the guidance effect. Then a family of bell-shaped, time-varying curves based on the beta distribution shifts the trade-off across the denoising steps. In tests the method improved image quality scores while holding text-image match near the standard approach, softening the trade-off rather than picking a side.

Catalog record

CREATORDawid Malarz, Artur Kasymov, Maciej Zieba, Jacek Tabor, Przemysław Spurek
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-10574

Browse the full library · About the Tower of Babel