From Scale to Speed: Adaptive Test-Time Scaling for Image Editing

Xiangyan Qu; Zhenlong Yuan; Jing Tang; Rui Chen; Datao Tang; Meng Yu; Lei Sun; Yancheng Bai

cs.CVFeb 24, 2026

From Scale to Speed: Adaptive Test-Time Scaling for Image Editing

Authors: Xiangyan Qu, Zhenlong Yuan, Jing Tang, Rui Chen, Datao Tang, Meng Yu, Lei Sun, Yancheng Bai, +4 more

Organizations: Institute of Information Engineering, Chinese Academy of Sciences · School of Cyber Security, University of Chinese Academy of Sciences · AMAP, Alibaba Group · University of Queensland

Abstract

Image Chain-of-Thought (Image-CoT) is a test-time scaling paradigm that improves image generation by extending inference time. Most Image-CoT methods focus on text-to-image (T2I) generation. Unlike T2I generation, image editing is goal-directed: the solution space is constrained by the source image and instruction. This mismatch causes three challenges when applying Image-CoT to editing: inefficient resource allocation with fixed sampling budgets, unreliable early-stage verification using general MLLM scores, and redundant edited results from large-scale sampling. To address this, we propose ADaptive Edit-CoT (ADE-CoT), an on-demand test-time scaling framework to enhance editing efficiency and performance. It incorporates three key strategies: (1) a difficulty-aware resource allocation that assigns dynamic budgets based on estimated edit difficulty; (2) edit-specific verification in early pruning that uses region localization and caption consistency to select promising candidates; and (3) depth-first opportunistic stopping, guided by an instance-specific verifier, that terminates when intent-aligned results are found. Extensive experiments on three SOTA editing models (Step1X-Edit, BAGEL, FLUX.1 Kontext) across three benchmarks show that ADE-CoT achieves superior performance-efficiency trade-offs. With comparable sampling budgets, ADE-CoT obtains better performance with more than 2x speedup over Best-of-N.

Figures & tables

Figure 1: Impact of Image-CoT on different generative tasks . (a) T2I generation is an open-ended task benefiting from large-scale sampling. (b) Image editing is a goal-directed task where outputs are constrained by the prompt and source image, leading to redundant correct outputs after large-scale sampling.

Figure 2: Why existing Image-CoT methods are suboptimal for editing. (a) Inefficient resource allocation : Fixed budgets waste computation on simple edits (high initial scores, red box ) that show minimal improvement. (b) Unreliable early-stage verification : 40% of samples with low early scores achieve high final scores, but are pruned incorrectly by general MLLM scores. (c) Redundant edited results : Large-scale sampling produces redundant correct outputs with identical best scores, though only one is sufficient.

Figure 3: Pipeline comparison of Image-CoT methods for editing . (a) Best-of-N employs a breadth-first search with a fixed sampling budget, producing all $N$ candidates before verification. (b) Previous pruning strategies improve efficiency by pruning with general MLLM scores. (c) Our ADE-CoT employs three strategies for improved quality and efficiency: difficulty-aware resource allocation to dynamically adjust the sampling budget ( orange , Sec. 3.1 ), edit-specific verification to identify promising candidates in early denoising stage ( blue , Sec. 3.2 ), and depth-first opportunistic stopping to terminate search when intent-aligned results are obtained ( green , Sec. 3.3 ).

Model	$N$	GEdit-Bench-EN [ 44 ] (Full set)					AnyEdit-Test [ 96 ]					Reason-Edit [ 26 ]
Model	$N$	G_SC	G_PQ	G_O	$\eta$	$\xi$	$\text{CLIP}_{\text{im}}$	$\text{CLIP}_{\text{out}}$	DINO	$\eta$	$\xi$	PSNR	LPIPS $\downarrow$	CLIP	$\eta$	$\xi$
FLUX.1 Kontext [ 34 ]	1	6.517	7.548	6.021	-	-	0.874	0.302	0.774	-	-	25.135	0.066	21.361	-	-
w/ BoN [ 48 ]	32	7.132	7.721	6.641	0.66	0.12	0.882	0.307	0.784	0.66	0.21	25.657	0.054	21.635	0.24	0.24
w/ PRM [ 104 ]	32	7.018	7.713	6.517	1.13	0.27	0.880	0.306	0.782	0.99	0.37	25.633	0.058	21.603	0.31	0.44
w/ PARM [ 104 ]	32	7.087	7.716	6.563	0.77	0.29	0.881	0.307	0.783	0.85	0.36	25.498	0.056	21.656	0.26	0.39
w/ TTS-EF [ 107 ]	32	6.866	7.657	6.376	0.98	0.57	0.878	0.305	0.779	0.72	0.33	25.509	0.057	21.639	0.44	0.46

Table 1: Comparison of ADE-CoT with SOTA Image-CoT methods .

\eta

measures performance-efficiency trade-off and

\xi

measures generation redundancy. Except for LPIPS, higher values are better for other metrics. The best results within a model are in bold . Non-degraded performances compared to the BoN method after pruning strategies are highlighted . All results are the average of three runs.

Panel of Figure 4 — Figure 4: Scaling curves on GEdit-Bench across different editing models. We show overall performance (G_O, y-axis) versus computational cost (NFE, x-axis) for sampling budgets of $N=1,2,4,8,16,32$ . The shaded regions indicate error bars. Our ADE-CoT ( purple star ) consistently surpasses SOTA Image-CoT methods across all models and budgets, achieving a better performance-efficiency trade-off.

Panel of Figure 5 — Figure 5: Effect of $\gamma$ in difficulty-aware resource allocation .

Panel of Figure 6 — Figure 6: Effect of edit-specific verification on early pruning.

Figure 7: Instance-specific verification detects subtle errors . Both candidates receive similarly high general scores. In contrast, our instance-specific verifier reveals the critical flaw in the left image (“ head turned sideways ”), improving final selection accuracy.

Panel of Figure 8 — Figure 8: Impact of opportunistic stopping strategy.

Model	Kontext		BAGEL		Step1X-Edit
Model	G_O $\uparrow$	NFE $\downarrow$	G_O $\uparrow$	NFE $\downarrow$	G_O $\uparrow$	NFE $\downarrow$
Baseline (BoN)	6.641	896	6.908	1600	7.157	896
a) $+$ difficulty-aware budgets	6.641	797	6.909	1391	7.157	778
b) $+$ early pruning (w/ $S_{\text{gen}}$ )	6.642	719	6.912	1351	7.157	719
c) $+$ early pruning (w/ $S$ )	6.647	673	6.916	1290	7.161	638
d) $+$ filtering similar samples	6.651	508	6.915	1087	7.162	522

Table 2: Effect of the three proposed strategies on efficiency and performance . We evaluate our method on GEdit-Bench [ 44 ] .

Model	Kontext		BAGEL		Step1X-Edit
Model	G_O $\uparrow$	NFE $\downarrow$	G_O $\uparrow$	NFE $\downarrow$	G_O $\uparrow$	NFE $\downarrow$
a) w/ additional steps	6.678	523	6.952	1008	7.188	525
b) w/ noisy latents	6.648	790	6.945	1334	7.153	765
ADE-CoT (full)	6.695	418	6.972	882	7.196	434

Table 3: Ablation of obtaining early preview images .

Model	Kontext			BAGEL			Step1X-Edit
Model	G_O $\uparrow$	NFE $\downarrow$	$\eta$ $\uparrow$	G_O $\uparrow$	NFE $\downarrow$	$\eta$ $\uparrow$	G_O $\uparrow$	NFE $\downarrow$	$\eta$ $\uparrow$
a) w/ BFS [ 55 ]	6.702	464	1.37	6.984	972	1.12	7.206	462	1.36
b) w/ DFS [ 78 ]	6.644	574	1.32	6.966	1073	1.08	7.162	554	1.29
c) w/o sorting	6.694	433	1.42	6.972	905	1.23	7.194	445	1.41
ADE-CoT (full)	6.695	418	1.47	6.972	882	1.27	7.196	434	1.45

Table 4: Ablation of different search ways for pruning strategy .

Panel of Figure 9 — Figure 9: Effect of timesteps in early prune and late retain .

Model	MLLM	Kontext		BAGEL		Step1X-Edit
Model	MLLM	G_O $\uparrow$	NFE $\downarrow$	G_O $\uparrow$	NFE $\downarrow$	G_O $\uparrow$	NFE $\downarrow$
BoN	Qwen2.5-VL-72B [ 3 ]	6.568	896	7.001	1600	7.155	896
ADE-CoT	Qwen2.5-VL-72B [ 3 ]	6.637	436	7.042	897	7.193	446
BoN	Qwen-VL-MAX [ 3 ]	6.641	896	6.908	1600	7.157	896
ADE-CoT	Qwen-VL-MAX [ 3 ]	6.695	418	6.972	882	7.196	434
BoN	Qwen3-VL-32B [ 93 ]	6.691	896	7.034	1600	7.158	896

Table 5: Effect of MLLMs on performance and efficiency .

Appendix figures & tables19 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 10: Qualitative comparison on complex edits. We compare three SOTA editing models (FLUX.1 Kontext [ 34 ] , BAGEL [ 15 ] , and Step1X-Edit [ 44 ] ) on challenging edits (large pose changes, multi-object modifications, and fine-grained regional edits). Baseline models often fail, while our ADE-CoT produces correct results via adaptive test-time scaling. Zoom in for detailed view.

Figure 11: Qualitative comparison on multi-turn edits . The baseline model often fails to preserve the context from previous edits, leading to accumulated errors in subsequent turns. Our ADE-CoT maintains consistency across multiple sequential instructions, producing correct final images that reflect all requested changes. Zoom in for detailed view.

Figure 12: Inefficient resource allocation with fixed sampling budgets . This figure extends the analysis from Fig. 2(a) to three SOTA models (FLUX.1 Kontext, BAGEL, and Step1X-Edit) and three benchmarks (GEdit-Bench, AnyEdit, and Reason-Edit). Edits with high initial scores ( red boxes , [8,9) and [9,10)) show minimal improvement across all three models and three benchmarks. Edits with low initial scores (indicating complex tasks) benefit significantly from Image-CoT, achieving substantial performance gains. This demonstrates that fixed sampling budgets waste computation on simple edits, motivating our difficulty-aware resource allocation strategy.

Figure 13: Misjudgement by general MLLM scores in early pruning. This figure extends the analysis from Fig. 2(b) to three SOTA editing models (FLUX.1 Kontext, BAGEL, and Step1X-Edit) evaluated on three benchmarks (GEdit-Bench, AnyEdit-Test, and Reason-Edit). Samples with low early scores are categorized by their final scores (x-axis): low score region ( red , [0,6)) and high score region ( green , [6,9)). On average, 37% of early low-scoring samples eventually achieve high final scores, yet would be incorrectly discarded by general MLLM scores. This demonstrates unreliable early-stage verification in editing, motivating our edit-specific verification strategy.

Figure 14: Redundant edited results in large-scale sampling. This figure extends the analysis from Fig. 2(c) to three SOTA editing models (FLUX.1 Kontext, BAGEL, and Step1X-Edit) evaluated on three benchmarks (GEdit-Bench, AnyEdit-Test, and Reason-Edit). For most edit cases, a large number of candidates (green bars, left y-axis) share identical best scores. The number of edit cases exhibiting such redundancy (orange bars, right y-axis) increases significantly in high score regions, particularly in [7,8) and [8,9) (x-axis). This demonstrates that Image-CoT produces redundant correct outputs in goal-directed editing, motivating our opportunistic stopping strategy.

Figure 15: Effectiveness of the one-step preview mechanism . We compare the noisy latent $I_{t}$ (top row for each model) with our corresponding one-step preview image $I_{0|t}$ (bottom row) at various denoising timesteps. Across all three models, our one-step preview generates a clear and high-fidelity approximation of the final output, even at very early stages ( e.g . , $t=8$ ). This demonstrates that the preview accurately reflects the final image’s content and quality, providing a solid basis for our edit-specific verifiers.

Panel of Figure 16 — Figure 16: Prompt $P_{\text{reg}}$ for identifying edited and keep objects .

Panel of Figure 17 — Figure 17: Prompt $P_{\text{cap}}$ for generating output caption .

Figure 18: Redundancy appears in early preview images. The figure displays clusters of visually similar images, grouped by colored boxes. We observe that candidates which are highly similar in the final output (bottom row) are also highly similar in their early previews (top row). This confirms that visual redundancy can be detected and filtered at an early stage.

Panel of Figure 19 — Figure 19: Prompt $P_{q}$ for generating instance-specific questions .

Panel of Figure 20 — Figure 20: Prompt $P_{a}$ for answering instance-specific questions .

Hyper.	Value	Description
$T$	28, 28, 50	Total denoising steps for each model.
$t_{e}$	8, 8, 16	Timestep for early preview and pruning.
$t_{l}$	16, 16, 36	Timestep for late retaining.
$N_{\text{min}}$	1	Minimum sampling budget.
$\gamma$	0.15	Sensitivity for difficulty-aware allocation.
$S_{\text{max}}$	10	Maximum score for normalization.

Appendix

Table 6: Default hyperparameters used in our experiments.

Model	$N$	FLUX.1 Kontext [ 34 ]				BAGEL [ 15 ]				Step1X-Edit [ 44 ]
Model	$N$	G_O $\uparrow$	NFE $\downarrow$	$\eta$ $\uparrow$	$\xi$ $\uparrow$	G_O $\uparrow$	NFE $\downarrow$	$\eta$ $\uparrow$	$\xi$ $\uparrow$	G_O $\uparrow$	NFE $\downarrow$	$\eta$ $\uparrow$	$\xi$ $\uparrow$
Baseline (BoN)	32	6.641	896	0.66	0.11	6.908	1600	0.69	0.14	7.157	896	0.72	0.13
a) $+$ adaptive sampling	32	6.641	797	0.74	0.26	6.909	1391	0.76	0.23	7.157	778	0.81	0.27
b) $+$ early filter by general verifier	32	6.642	719	0.81	0.40	6.912	1351	0.79	0.39	7.157	719	0.89	0.42
c) $+$ early filter (+ $S_{\text{reg}}$ )	32	6.645	687	0.84	0.42	6.915	1321	0.84	0.43	7.158	678	0.95	0.45
d) $+$ early filter (+ $S_{\text{cap}}$ )	32	6.647	673	0.87	0.44	6.916	1290	0.88	0.45	7.161	638	1.02	0.47

Appendix

Table 7: Effect of the three proposed strategies on efficiency and performance . We evaluate our method on GEdit-Bench [ 44 ] .

Panel of Figure 21 — Figure 21: Correlation between preview scores and final performance across denoising stages. We show correlation matrixs between intermediate scores (y-axis) and final scores (x-axis) at (a) the early and (b) late denoising stages.

Panel of Figure 22 — Figure 22: More hyperparameter analysis .

MLLM Component	Kontext	BAGEL	Step1X-Edit
General Verifier	65.1	71.4	66.1
Region Generation	1.0	1.0	1.0
Caption Generation	1.0	1.0	1.0
Instance-Specific Questions (Generate)	1.0	1.0	1.0
Instance-Specific Questions (Answer)	14.9	17.6	15.5
Total Avg.	83.0	92.0	84.6

Appendix

Table 8: Average MLLM queries per editing case on GEdit-Bench . We report the average number of calls for each MLLM component under a sampling budget of

N=32

Figure 23: Instance-specific verifier improves fine-grained selection . We compare the baseline model, Best-of-N (BoN), and our ADE-CoT across challenging editing scenarios. Baseline models and BoN often produce results with subtle errors (marked in red boxes), such as incorrect pose changes, background adjustments, and clothing modifications. In contrast, our instance-specific verifier generates targeted questions to examine critical details, enabling accurate detection of editing errors and selecting correct results (marked in green boxes).

Model	MLLM for Region	Kontext		BAGEL		Step1X-Edit
Model	MLLM for Region	G_O $\uparrow$	NFE $\downarrow$	G_O $\uparrow$	NFE $\downarrow$	G_O $\uparrow$	NFE $\downarrow$
BoN	-	6.641	896	6.908	1600	7.157	896
ADE-CoT	Qwen2.5-VL-72B [ 3 ]	6.637	436	7.042	897	7.193	446
	Qwen-VL-MAX [ 3 ]	6.661	428	6.993	901	7.194	445
	Qwen3-VL-32B [ 93 ]	6.673	424	7.048	869	7.198	436

Appendix

Table 9: Effect of MLLMs on edited-region correctness . We compare three MLLMs for edited-region localization (via prompt

P_{\text{reg}}

), while keeping Qwen-VL-72B for other components.

Model	MLLM for Caption	Kontext		BAGEL		Step1X-Edit
Model	MLLM for Caption	G_O $\uparrow$	NFE $\downarrow$	G_O $\uparrow$	NFE $\downarrow$	G_O $\uparrow$	NFE $\downarrow$
BoN	-	6.641	896	6.908	1600	7.157	896
ADE-CoT	Qwen2.5-VL-72B [ 3 ]	6.637	436	7.042	897	7.193	446
	Qwen-VL-MAX [ 3 ]	6.651	419	7.021	899	7.186	450
	Qwen3-VL-32B [ 93 ]	6.664	423	7.052	883	7.195	440

Appendix

Table 10: Effect of MLLMs on instruction-caption consistency . We compare three MLLMs for instruction-caption consistency (via prompt

P_{\text{cap}}

), while keeping Qwen-VL-72B for others.

Explore similar work

Adaptive Inference-Time Scaling via Early-Step Latent Verification for Image Editing
Jun 13, 2026Yue Yu, Yang Jiao, Jiayu Wang +2Inference-Time Scaling Diffusion-Based Image Editing
College of Computer Science and Artificial Intelligence, Fudan University, Shanghai 200433, China · Microsoft Research Asia, Beijing 100080, China · Institute of Trustworthy Embodied AI, Fudan University, Shanghai 200433, China
Meta-CoT: Enhancing Granularity and Generalization in Image Editing
Apr 27, 2026Shiyi Zhang, Yiji Cheng, Tiankai Hang +8Text-Guided Image Editing Unified Multimodal Models
Shenzhen International Graduate School, Tsinghua University · Hunyuan, Tencent
Making Image Editing Easier via Adaptive Task Reformulation with Agentic Executions
Apr 17, 2026Bo Zhao, Kairui Guo, Runnan Du +6Text-Guided Image Editing
Nanjing University China · Beijing Institute of Technology China · College of Artificial Intelligence, China University of Petroleum (Beijing) China +3

Jun 13, 2026J/K move · Enter open · S save

Adaptive Inference-Time Scaling via Early-Step Latent Verification for Image Editing

Yue Yu, Yang Jiao, Jiayu Wang, Qi Dai+1

College of Computer Science and Artificial Intelligence, Fudan University, Shanghai 200433, China · Microsoft Research Asia, Beijing 100080, China · Institute of Trustworthy Embodied AI, Fudan University, Shanghai 200433, China

Inference-Time Scaling Diffusion-Based Image Editing Diffusion Model Inference Acceleration

Open paper AskPDF arXiv

Instruction-based image editing has made notable progress with recent advances in generative models. However, the quality of the edited result is still influenced by the randomly sampled initial noise, particularly in complex editing scenarios. An unsuitable initial noise may lead to unsatisfactory editing results. Recent inference-time scaling methods address this issue by sampling multiple initial noises and selecting better candidates. Nevertheless, most of them follow a decode-then-verify scheme which introduces an efficiency-accuracy trade-off. When decoding is performed after limited inference steps, the decoded images often remain too noisy for reliable assessment, whereas sufficiently denoised images require much higher computational cost. To address this issue, we propose VeriLatent, a plug-and-play adaptive inference-time scaling framework with early-step latent verification for image editing. Specifically, we propose a novel verifier that scores each initial noise through a latent-space editing activation map at an early stage. It identifies promising candidates by assessing whether they can induce an effective edit in the correct region. This enables efficient early pruning without decoding latents into images. Building on this, we further develop an adaptive search strategy for inference-time scaling. It allocates inference budgets according to editing difficulty, thereby reducing the number of function evaluations (NFE). Extensive experiments on multiple benchmarks and different base models demonstrate that VeriLatent consistently improves both editing performance and inference-time scaling efficiency.

Abstract

Explore similar work

Adaptive Inference-Time Scaling via Early-Step Latent Verification for Image Editing

Meta-CoT: Enhancing Granularity and Generalization in Image Editing

Making Image Editing Easier via Adaptive Task Reformulation with Agentic Executions