CoEditor++: Instruction-based Visual Editing via Cognitive Reasoning

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Ni, Minheng, Fan, Yutao, Yang, Zhengyuan, Shen, Yeli, Wei, Yuxiang, Zhang, Yaowen, Wang, Lijuan, Zhang, Lei, Zuo, Wangmeng
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910042776666112
author Ni, Minheng
Fan, Yutao
Yang, Zhengyuan
Shen, Yeli
Wei, Yuxiang
Zhang, Yaowen
Wang, Lijuan
Zhang, Lei
Zuo, Wangmeng
author_facet Ni, Minheng
Fan, Yutao
Yang, Zhengyuan
Shen, Yeli
Wei, Yuxiang
Zhang, Yaowen
Wang, Lijuan
Zhang, Lei
Zuo, Wangmeng
contents Recent advances in large multimodal models (LMMs) have enabled instruction-based image editing, allowing users to modify visual content via natural language descriptions. However, existing approaches often struggle with high-level semantic reasoning and visual consistency, particularly under ambiguous or complex instructions. To address these challenges, we propose CoEditor++, a cognitively structured, training-free framework that decomposes editing into "what to edit" and "how to edit" through two cognitive stages with a reflective self-selection mechanism, enabling robust, fine-grained, and interpretable editing. Built entirely from open-sourced components, CoEditor++ requires no additional training or fine-tuning, ensuring transparency and cross-domain applicability. We evaluate CoEditor++ on SmartEdit, a widely used benchmark for general editing, and AltBear, a privacy and compliance-oriented benchmark. Experimental results show that CoEditor++ achieves state-of-the-art performance in both general editing and responsible editing tasks compared with open-sourced models that require training on specialized editing datasets maintaining significantly higher visual consistency. When compared with closed-source models such as Nano Banana Pro or GPT-4o, CoEditor++ preserves comparable instruction following while still substantially outperforming them in visual consistency. Extensive ablation studies confirm that the effectiveness of CoEditor++ benefits from its structured cognitive design rather than any specific model component. Our findings suggest the potential toward cognitive-centric instruction-based image editing.
format Preprint
id arxiv_https___arxiv_org_abs_2603_05518
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle CoEditor++: Instruction-based Visual Editing via Cognitive Reasoning
Ni, Minheng
Fan, Yutao
Yang, Zhengyuan
Shen, Yeli
Wei, Yuxiang
Zhang, Yaowen
Wang, Lijuan
Zhang, Lei
Zuo, Wangmeng
Human-Computer Interaction
Computer Vision and Pattern Recognition
Recent advances in large multimodal models (LMMs) have enabled instruction-based image editing, allowing users to modify visual content via natural language descriptions. However, existing approaches often struggle with high-level semantic reasoning and visual consistency, particularly under ambiguous or complex instructions. To address these challenges, we propose CoEditor++, a cognitively structured, training-free framework that decomposes editing into "what to edit" and "how to edit" through two cognitive stages with a reflective self-selection mechanism, enabling robust, fine-grained, and interpretable editing. Built entirely from open-sourced components, CoEditor++ requires no additional training or fine-tuning, ensuring transparency and cross-domain applicability. We evaluate CoEditor++ on SmartEdit, a widely used benchmark for general editing, and AltBear, a privacy and compliance-oriented benchmark. Experimental results show that CoEditor++ achieves state-of-the-art performance in both general editing and responsible editing tasks compared with open-sourced models that require training on specialized editing datasets maintaining significantly higher visual consistency. When compared with closed-source models such as Nano Banana Pro or GPT-4o, CoEditor++ preserves comparable instruction following while still substantially outperforming them in visual consistency. Extensive ablation studies confirm that the effectiveness of CoEditor++ benefits from its structured cognitive design rather than any specific model component. Our findings suggest the potential toward cognitive-centric instruction-based image editing.
title CoEditor++: Instruction-based Visual Editing via Cognitive Reasoning
topic Human-Computer Interaction
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.05518