Generative Visual Chain-of-Thought for Image Editing

Zijin Yin; Tiankai Hang; Yiji Cheng; Shiyi Zhang; Runze He; Yu Xu; Chunyu Wang; Bing Li

cs.CVMar 2, 2026

Generative Visual Chain-of-Thought for Image Editing

Authors: Zijin Yin, Tiankai Hang, Yiji Cheng, Shiyi Zhang, Runze He, Yu Xu, Chunyu Wang, Bing Li,

Organizations: Beijing University of Posts and Telecommunications · Beijing Key Laboratory of Multimodal Data Intelligent Perception and Governance · Tencent Hunyuan · King Abdullah University of Science and Technology

Abstract

Existing image editing methods struggle to perceive where to edit, especially under complex scenes and nuanced spatial instructions. To address this issue, we propose Generative Visual Chain-of-Thought (GVCoT), a unified framework that performs native visual reasoning by first generating spatial cues to localize the target region and then executing the edit. Unlike prior text-only CoT or tool-dependent visual CoT paradigms, GVCoT jointly optimizes visual tokens generated during the reasoning and editing phases in an end-to-end manner. This way fosters the emergence of innate spatial reasoning ability and enables more effective utilization of visual-domain cues. The main challenge of training GCVoT lies in the scarcity of large-scale editing data with precise edit region annotations; to this end, we construct GVCoT-Edit-Instruct, a dataset of 1.8M high-quality samples spanning 19 tasks. We adopt a progressive training strategy: supervised fine-tuning to build foundational localization ability in reasoning trace before final editing, followed by reinforcement learning to further improve reasoning and editing quality. Finally, we introduce SREdit-Bench, a new benchmark designed to comprehensively stress-test models under sophisticated scenes and fine-grained referring expressions. Experiments demonstrate that GVCoT consistently outperforms state-of-the-art models on SREdit-Bench and ImgEdit. We hope our GVCoT will inspire future research toward interpretable and precise image editing.

Figures & tables

Figure 2: Comparing spatial cue representation for image editing on ImgEdit [ 68 ] . We study two ways of injecting spatial information: (1) text modality uses bounding box coordinates, and (2) visual modality providing a binary mask. Providing spatial information in the visual modality yields a greater improvement in both instruction adherence and background preservation.

Figure 3: Supervised Fine-Tuning of our GVCoT training recipe. Stage 1: Multi-Task Visual Manipulation, where the model’s generation expert is trained in a multi-task setup to inject the newly masking skill. Stage 2: Visual Reason-aided Editing, where the entire model is trained to generate a faithful and interpretable visual reasoning image and then an edited image within a single sequence.

Figure 4: GVCoT-Edit-Instruct Data Pipeline. Left: We design a scalable multi-stage data pipeline to curate high-quality samples with faithful editing region annotations, i.e. , bounding boxes and masks. Right: The distribution of GVCoT-Edit-Instruct spanning 19 tasks.

Figure 5: Illustration of the SREdit-Bench . Left: We provide challenging scenarios featuring complex scenes and fine-grained referring expressions. Right: (a) We quantify scene complexity by counting editable objects and regions. Results show that SpaEdit-Bench concentrates on more sophisticated scenes than ImgEdit [ 68 ] and GEdit-Bench [ 40 ] . (b) Referral type distribution. (c) Edit tasks distribution.

Product-level models
Model	SREdit-Bench $\uparrow$
Model	$\text{SC}_{g}$	$\text{PQ}_{g}$	$\text{O}_{g}$
GPT Image 1 [High] [ 47 ]	9.02	8.42	8.56
FLUX.1 Kontext [Pro]	8.69	8.40	8.41
Qwen-Image [ 63 ]	8.57	8.43	8.32
Generation Only

Table 1: Quantitative results on SREdit-Bench.

\text{SC}_{g}

\text{PQ}_{g}

, and

\text{O}_{g}

indicate scores on Semantic Consistency, Perceptual Quality, and Overall. The best and second-best results are highlighted in bold and underlined, respectively.

Method	CLIP-I $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$
SmartEdit [ 24 ]	0.8841	0.2915	17.1728	0.6828
BrushNet [ 28 ]	0.8986	0.1830	19.2172	0.7877
MagicQuill [ 42 ]	0.9381	0.1162	22.2380	0.8981
MIND-Edit [ 61 ]	0.9310	0.1245	22.2714	0.8517
Bagel [ 12 ]	0.9124	0.1721	22.1843	0.8640
Bagel-GVCoT (Ours)	0.9451	0.1066	23.0161	0.8943

Table 2: Quantitative results in HumanEdit. Our Bagel-GVCoT allowing native visual reasoning outperforms previous approaches that rely on additional mask guidance.

Product-level models
Model	Add	Adjust	Extract	Replace	Remove	Background	Style	Hybrid	Action	Overall $\uparrow$
FLUX.1 Kontext [Pro] [ 31 ]	4.25	4.15	2.35	4.56	3.57	4.26	4.57	3.68	4.63	4.00
GPT Image 1 [High] [ 47 ]	4.61	4.33	2.90	4.35	3.66	4.57	4.93	3.96	4.89	4.20
Qwen-Image [ 63 ]	4.38	4.16	3.43	4.66	4.14	4.38	4.81	3.82	4.69	4.27
Generation Only
MagicBrush [ 70 ]	2.84	1.58	1.51	1.97	1.58	1.75	2.38	1.62	1.22	1.90

Table 3: Quantitative results on ImgEdit. We use GPT-4.1 to evaluate all metrics. “Overall” is calculated by averaging all scores across tasks. The best and second-best results in open-sourced models are highlighted in bold and underlined , respectively.

Figure 6: Quantitative comparison of different visual reasoning paradigms on SREdit-Bench. The performance is measured by

\text{O}_{g}

. Under both visual cue forms, our method consistently surpasses the text CoT and Visual CoT with considerable margins.

Figure 7: Qualitative comparison in SREdit-Bench. Our method demonstrates superior spatial reasoning and instruction adherence compared to existing open-source models, especially when handling complex, multi-object editing tasks.

	Model	IoU $\uparrow$	$\text{SC}_{g}$ $\uparrow$	$\text{PQ}_{g}$ $\uparrow$	$\text{O}_{g}$ $\uparrow$
	Bagel *	–	6.21	6.09	5.71
	TCoT	–	6.54	6.22	5.97
Mask	VCoT	0.68	7.55	7.92	7.31
	GVCoT	0.60	8.53	8.62	8.42
	VCoT w/ TF-thought	–	8.03	7.96	7.75
	GVCoT w/ TF-thought	–	8.79	8.95	8.72

Table 4: Comparative results of GVCoT and Visual CoT using SREdit-Bench. TF-thought denotes skipping the reasoning step by using teacher-forcing visual thought.

Step 1	Step 2	IoU $\uparrow$	$\text{SC}_{g}$ $\uparrow$	$\text{PQ}_{g}$ $\uparrow$	$\text{O}_{g}$ $\uparrow$
$\checkmark$		0.64	7.92	8.01	7.75
	$\checkmark$	0.61	8.21	8.33	8.10
$\checkmark$	$\checkmark$	0.66	8.53	8.62	8.42

Table 5: Ablation results on progressive SFT.

(1) Multi-stage RL training
Setting	IoU $\uparrow$	$\text{SC}_{g}$ $\uparrow$	$\text{PQ}_{g}$ $\uparrow$	$\text{O}_{g}$ $\uparrow$
SFT Only	0.60	8.53	8.62	8.42
SFT+RL	0.67	8.57	8.76	8.53
w/o Stage 2	0.67	8.32	8.47	8.25
(2) Visual thought reward design
Full reward set	0.67	8.32	8.47	8.25

Table 6: Ablation studies on Reinforcement Learning (RL). We analyze (1) multi-stage training strategy, (2) visual-thought reward design, and (3) editing reward design.

Appendix figures & tables15 assets

Supplementary material from the paper’s appendix.

Appendix

Product-level models
Model	GEdit-Bench-EN (Full set)
Model	G_SC $\uparrow$	G_PQ $\uparrow$	G_O $\uparrow$
FLUX.1 Kontext [Pro] [ 31 ]	7.02	7.60	6.56
Gemini 2.0 [ 10 ]	6.73	6.61	6.32
Gemini 2.5 Flash Image Preview [ 11 ]	7.28	7.83	6.93
Gemini 2.5 Flash Image [ 11 ]	7.41	7.96	7.10

Appendix

Table 1: Quantitative results on GEdit-Bench We use Semantic Consistency (G_SC), Perceptual Quality (G_PQ), and Overall Score (G_O) metrics measured by GPT 4.1.

Figure 1: Limitation of our Bagel-GVCoT. Our approach constrains under style transfer tasks compared with the baseline.

	Emu Edit Test set					MagicBrush Test Set
Method	$\text{CLIP}_{dir}\!\uparrow$	$\text{CLIP}_{im}\!\uparrow$	$\text{CLIP}_{out}\!\uparrow$	$\text{L1}\!\downarrow$	DINO $\uparrow$	$\text{CLIP}_{dir}\!\uparrow$	$\text{CLIP}_{im}\!\uparrow$	$\text{CLIP}_{out}\!\uparrow$	$\text{L1}\!\downarrow$	DINO $\uparrow$
InstructPix2Pix [ 4 ]	0.078	0.834	0.219	0.121	0.762	0.115	0.837	0.245	0.093	0.767
MagicBrush [ 70 ]	0.090	0.838	0.222	0.100	0.776	0.123	0.883	0.261	0.058	0.871
PnP [ 55 ]	0.028	0.521	0.089	0.304	0.153	0.025	0.568	0.101	0.289	0.220
Null-Text Inv. [ 45 ]	0.101	0.761	0.236	0.075	0.678	0.121	0.752	0.263	0.077	0.664
UltraEdit [ 73 ]	0.107	0.793	0.283	0.071	0.844	-	0.868	-	0.088	0.792

Appendix

Table 2: Quantitative results on Emu Edit and MagicBrush test sets. All results except Bagel [ 12 ] and our Bagel-GVCoT are directly from Unireal [ 7 ] . The best and second-best results are highlighted in bold and underlined , respectively.

Figure 2: Statistical analysis of GVCoT-Edit-Instruct. Left: Length distribution of textual editing instructions. Middle: Word cloud of all editing instructions. Right: Proportion of edit-region masks that cover the entire image.

Figure 3: User study. We conduct GSB (Good/Same/Bad) to compare our method with state-of-the-art models.

Figure 4: More examples of our training dataset GVCoT-Edit-Instruct. Each sample consists of four components: (1) the source image, (2) the textual edit instruction, (3) the edit mask, a pink overlay indicating the edit region, and (4) the target image after editing.

Figure 5: More examples of SREdit-Bench. We focus on fine-grained referring expressions and sophisticated visual scenes.

Optimizer & Scheduler
Hyperparameter	Stage 1	Stage 2
Learning Rate (LR)	$2\times 10^{-5}$	$1\times 10^{-5}$
LR Scheduler	Cosine Decay	Cosine Decay
Min Learning Rate	$1\times 10^{-7}$	$1\times 10^{-7}$
Warmup Steps	1,000	1,000
Total Training Steps	80,00	12,000

Appendix

Table 3: Key hyperparameters for the two-stage training process in our Progressive Supervised Fine-Tuning. “ N/A ” indicates that the parameter did not apply to that stage.

Optimizer & Scheduler
Hyperparameter	Stage 1	Stage 2
Learning Rate (LR)	$1\times 10^{-6}$	$1\times 10^{-6}$
LR Scheduler	Cosine Decay	Cosine Decay
Min Learning Rate	$1\times 10^{-7}$	$1\times 10^{-7}$
Warmup Steps	50	50
Total Training Steps	500	450

Appendix

Table 4: Key hyperparameters for the two-stage training process in our Reinforcement-based Refining. “ N/A ” indicates that the parameter did not apply to that stage.

Figure 6: Qualitative comparison in SREdit-Bench. Our method demonstrates superior spatial reasoning and instruction adherence compared to existing open-source models, especially when handling complex, multi-object editing tasks.

Panel of Table 5 — Table 5: The prompt template used by Qwen2.5-VL-72B [ 2 ] to measure the consistency of the edited image with CoT visual thought.

Panel of Table 6 — Table 6: The prompt template used by Qwen2.5-VL-72B [ 2 ] to measure the perceptual quality of source and edited images.

Panel of Table 7 — Table 7: The prompt template used by Qwen2.5-VL-72B [ 2 ] for generating diverse image editing instructions.

Panel of Table 8 — Table 8: The prompt template used by Qwen2.5-VL-72B [ 2 ] to measure the semantic consistency of triplets, i.e. , source image, edited image, and edit instruction, during our data curation procedure.

Panel of Table 9 — Table 9: The prompt template used by Qwen2.5-VL-72B [ 2 ] for mining bounding box coordinates of the edit region.

Text-Guided Image Editing

Text-guided image editing changes selected image content from natural-language instructions while aiming to preserve everything else.

See all 159 papers

InterCoG: Towards Spatially Precise Image Editing with Interleaved Chain-of-Grounding Reasoning
Mar 2, 2026Yecong Wan, Fan Li, Chunwei Wang +3Image Editing Text-Guided Image Editing
Faculty of Computing, Harbin Institute of Technology, Harbin, 150001, China. · Zhengzhou Advanced Research Institute of Harbin Institute of Technology, Zhengzhou, 450000, China. · Huawei Noah’s Ark Lab, Shenzhen, 518100, China. +1
Text-Vision Co-Instructed Image Editing
Jun 15, 2026Chenxi Xie, Yuhui Wu, Qiaosi Yi +1Text-Guided Image Editing Image Editing
The Hong Kong Polytechnic University · OPPO Research Institute
Instruction-based Image Editing with Planning, Reasoning, and Generation
Feb 26, 2026Liya Ji, Chenyang Qi, Qifeng ChenText-Guided Image Editing Diffusion-Based Image Editing
HKUST

Mar 2, 2026J/K move · Enter open · S save

InterCoG: Towards Spatially Precise Image Editing with Interleaved Chain-of-Grounding Reasoning

Yecong Wan, Fan Li, Chunwei Wang, Hao Wu+2

Faculty of Computing, Harbin Institute of Technology, Harbin, 150001, China. · Zhengzhou Advanced Research Institute of Harbin Institute of Technology, Zhengzhou, 450000, China. · Huawei Noah’s Ark Lab, Shenzhen, 518100, China.+1

Image Editing Text-Guided Image Editing Multimodal Grounding Visual Spatial Reasoning

Open paper AskPDF arXiv

Emerging unified editing models have demonstrated strong capabilities in general object editing tasks. However, it remains a significant challenge to perform fine-grained editing in complex multi-entity scenes, particularly those where targets are not visually salient and require spatial reasoning. To this end, we propose InterCoG, a novel text-vision Interleaved Chain-of-Grounding reasoning framework for fine-grained image editing in complex real-world scenes. The key insight of InterCoG is to first perform object position reasoning solely within text that includes spatial relation details to explicitly deduce the location and identity of the edited target. It then conducts visual grounding via highlighting the editing targets with generated bounding boxes and masks in pixel space, and finally rewrites the editing description to specify the intended outcomes. To further facilitate this paradigm, we propose two auxiliary training modules: multimodal grounding reconstruction supervision and multimodal grounding reasoning alignment to enforce spatial localization accuracy and reasoning interpretability, respectively. We also construct GroundEdit-45K, a dataset comprising 45K grounding-oriented editing samples with detailed reasoning annotations, and GroundEdit-Bench for grounding-aware editing evaluation. Extensive experiments substantiate the superiority of our approach in highly precise edits under spatially intricate and multi-entity scenes.

Abstract

Text-Guided Image Editing

Explore similar work

InterCoG: Towards Spatially Precise Image Editing with Interleaved Chain-of-Grounding Reasoning

Text-Vision Co-Instructed Image Editing

Instruction-based Image Editing with Planning, Reasoning, and Generation