Prompt-Guided Image Editing with Masked Logit Nudging in Visual Autoregressive Models

Amir El-Ghoussani, Marc Hölle, Gustavo Carneiro, Vasileios Belagiannis, [6pt] Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany · 2026

Read Prompt-Guided Image Editing with Masked Logit Nudging in Visual Autoregressive Models by Amir El-Ghoussani, Marc Hölle, Gustavo Carneiro, Vasileios Belagiannis, [6pt] Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany (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-2604-14591.txt.

Description

El-Ghoussani and three colleagues address prompt-guided editing in visual autoregressive models, changing a source image per a target prompt while keeping others intact. Masked Logit Nudging converts source encodings to logits and nudges predicted logits to targets along a semantic path from the source-target prompts. Edits apply only inside masks from cross-attention differences between prompts, with refinement fixing quantization errors. It tops PIE at 512 and 1024 pixels, reconstructs faithfully and beats prior methods on COCO and OpenImages, and runs much faster than diffusion models.

Catalog record

CREATORAmir El-Ghoussani, Marc Hölle, Gustavo Carneiro, Vasileios Belagiannis, [6pt] Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE73144
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.14591

Browse the full library · About the Tower of Babel