FireRed-Image-Edit-1.0 Technical Report

Super Intelligence Team, Xiaohongshu Inc · 2026

Read FireRed-Image-Edit-1.0 Technical Report by Super Intelligence Team, Xiaohongshu Inc (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-2602-13344.txt.

Description

Editing an image from a written instruction demands a model that keeps identity, follows wording, and handles text inside pictures. FireRed-Image-Edit is a diffusion transformer built for that job, trained through data curation, multi-stage training, and careful benchmark design. Its corpus began at 1.6 billion samples, then cleaning and two-stage filtering kept over 100 million high-quality pairs balanced between generation and editing.

Training moves from pretraining through fine-tuning to reinforcement learning, with new tricks for batching, prompt alignment, gradient stability, and text-aware rewards. A new benchmark spanning 15 editing types shows competitive or leading results against open and proprietary rivals, and code, models, and benchmarks are public. The report comes from the Super Intelligence Team, with 18 named researchers behind it.

Catalog record

CREATORSuper Intelligence Team, Xiaohongshu Inc
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE85 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-13344

Browse the full library · About the Tower of Babel