Discrete Noise Inversion for Next-scale Autoregressive Text-based Image Editing

Quan Dao, Xiaoxiao He, Ligong Han, Ngan Hoai Nguyen, Amin Heyrani Nobari, Faez Ahmed, Han Zhang, Viet Anh Nguyen, Dimitris Metaxas · 2025

Read Discrete Noise Inversion for Next-scale Autoregressive Text-based Image Editing by Quan Dao, Xiaoxiao He, Ligong Han, Ngan Hoai Nguyen, Amin Heyrani Nobari, Faez Ahmed, Han Zhang, Viet Anh Nguyen, Dimitris Metaxas (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-2509-01984.txt.

Description

Quan Dao leads eight co-authors with VARIN, the first noise inversion editor for visual autoregressive image models. Autoregressive models build images scale by scale, rivaling diffusion models on text to image tasks. Editing a given image by prompt, without retraining, is just as vital for real use. Inversion means running generation backward to recover the noise that rebuilds the source image.

Their key piece inverts the argmax sampling step with location awareness to yield inverse random noises. Those noises rebuild the source faithfully, then guide targeted edits from new prompts. Tests show strong prompt fidelity while backgrounds and structure stay intact. The method gives autoregressive models a practical editing path they lacked.

Catalog record

CREATORQuan Dao, Xiaoxiao He, Ligong Han, Ngan Hoai Nguyen, Amin Heyrani Nobari, Faez Ahmed, Han Zhang, Viet Anh Nguyen, Dimitris Metaxas
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-01984

Browse the full library · About the Tower of Babel