Better Source, Better Flow: Learning Condition-Dependent Source Distribution for Flow Matching

Junwan Kim, Jiho Park, Seonghu Jeon, Seungryong Kim · 2026

Read Better Source, Better Flow: Learning Condition-Dependent Source Distribution for Flow Matching by Junwan Kim, Jiho Park, Seonghu Jeon, Seungryong Kim (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-2602-05951.txt.

Description

Flow matching generators start from a source distribution and flow toward data. Nearly all of them inherit the standard Gaussian source from diffusion models, without asking whether it is a good start. Junwan Kim, Jiho Park, Seonghu Jeon, and Seungryong Kim propose learning a condition dependent source distribution instead. In text to image systems, rich conditioning signals are available before generation even begins.

Directly wiring conditioning into the source causes failure modes, including collapse and instability. The authors show that variance regularization and directional alignment between source and target are the critical stabilizers. They also analyze how the choice of target representation space changes where structured sources help most. Across text to image benchmarks, the method improved results consistently and converged far faster on image quality metrics. The work reframes the source distribution as a design choice, not a fixed default.

Catalog record

CREATORJunwan Kim, Jiho Park, Seonghu Jeon, Seungryong Kim
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE57 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-05951

Browse the full library · About the Tower of Babel