Post-Hoc Guidance for Consistency Models by Joint Flow Distribution Learning

Chia-Hong Hsu, Randall Balestriero · 2026

Read Post-Hoc Guidance for Consistency Models by Joint Flow Distribution Learning by Chia-Hong Hsu, Randall Balestriero (2026) 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-2604-08828.txt.

Description

Hsu and Balestriero propose Joint Flow Distribution Learning, a lightweight matching method bringing guidance without a classifier to ready made steady models without a spread teacher. Steady models sample in one or few steps, but current guidance needs copying from another spread teacher, capping guidance to copying methods. With the steady model as a plain changing formula solver, bell curve tests confirm the implied variance growing noise is Gaussian. The method unlocks guided making on a steady trained model while cutting FID on CIFAR-10 and ImageNet 64x64.

Catalog record

CREATORChia-Hong Hsu, Randall Balestriero
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64569
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.08828

Browse the full library · About the Tower of Babel