Match & Choose: Model Selection Framework for Fine-tuning Text-to-Image Diffusion Models

Basile Lewandowski, Lydia Y. Chen, Robert Birke · 2025

Read Match & Choose: Model Selection Framework for Fine-tuning Text-to-Image Diffusion Models by Basile Lewandowski, Lydia Y. Chen, Robert Birke (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-2508-10993.txt.

Description

Basile Lewandowski and co-authors tackle a practical choice: which public text-to-image model should a user fine-tune for their own data? Platforms host many pretrained diffusion models, and trying them all is costly. Their 2025 paper proposes Match and Choose, the first selection framework built for this setting.

The core is a matching graph. Its nodes are the candidate models and profiled datasets, and its edges score how models pair with data and how datasets resemble each other. From those scores the system ranks which model is most worth fine-tuning, sparing users the full sweep. Image generation teams get a shortcut that respects how domain fit, not raw size, decides fine-tuning success.

Catalog record

CREATORBasile Lewandowski, Lydia Y. Chen, Robert Birke
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE47 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-10993

Browse the full library · About the Tower of Babel