DIAMOND: Directed Inference for Artifact Mitigation in Flow Matching Models

Alicja Polowczyk, Agnieszka Polowczyk, Piotr Borycki, Joanna Waczyńska, Jacek Tabor, Przemysław Spurek · 2026

Read DIAMOND: Directed Inference for Artifact Mitigation in Flow Matching Models by Alicja Polowczyk, Agnieszka Polowczyk, Piotr Borycki, Joanna Waczyńska, Jacek Tabor, Przemysław Spurek (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-2602-00883.txt.

Description

Alicja Polowczyk and colleagues introduce DIAMOND, a training-free method that fixes paths during inference to reduce visual and body artifacts in text-to-image models. Current artifact fixes work post-hoc and miss the core image making process, or need invasive weight changes or costly refinement. DIAMOND rebuilds a clean-sample estimate at every step, steering output away from latent states that cause artifacts. Used with standard Diffusion Models, it offers a robust, zero-shot path to high-quality, artifact-free synthesis with no extra training.

Catalog record

CREATORAlicja Polowczyk, Agnieszka Polowczyk, Piotr Borycki, Joanna Waczyńska, Jacek Tabor, Przemysław Spurek
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE57080
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00883

Browse the full library · About the Tower of Babel