Scalable Minimal-Change Learning for Controllable Image Editing

Shuo Chen; Fengming Huang; Yu Yao; Mingming Gong; Tongliang Liu

cs.CVOct 5, 2026

Scalable Minimal-Change Learning for Controllable Image Editing

Authors: Shuo Chen, Fengming Huang, Yu Yao, Mingming Gong, Tongliang Liu

Organizations: Sydney AI Centre, The University of Sydney

Abstract

Image editing should change only the attributes specified by an instruction while preserving everything else, yet current methods often make unintended changes. We treat this minimal-change principle as an optimization objective for instruction-based editing. Latent L1 regularization is a poor proxy for output locality in modern nonlinear generators and often requires supervision unavailable at scale. We instead optimize edit outcomes with reinforcement learning. An agentic vision-language reward model audits each source image, instruction, and edited image for two failure types: unimplemented requested changes and unintended changes. A group-level rubric merges and verifies these issues to provide consistent rewards across candidate edits without per-instruction human annotations. On FLUX.1 Kontext-dev, ARRO raises average EditScore from 5.21 to 5.88 across MinEval, MagicBrush, AnyBench, and Emu-Edit. On 600 evaluation examples, it reduces off-target pixel change by 8.4% relative to the base editor. Reward and SFT controls, blinded human evaluations, and transfer to OmniGen2 provide complementary evidence. Code: https://github.com/Showwwwwwwww/ARRO

Figures & tables

Figure 1: Illustrative edits by FLUX.1 Kontext-dev and Arro . ARRO better preserves the companion, fruit arrangement, umbrella scene, and bus surroundings while carrying out the requested edits. The ambiguous “Include a delicious cake” example remains imperfect: ARRO adds a cake but replaces the original dish. It is retained as a visible failure case rather than evidence of perfect preservation.

Figure 2: Overview of our agentic reward pipeline. For each editing triplet $(\bm{x}_{\text{src}},c,\bm{x}_{i})$ , the vision-language model $\Phi_{\text{V}}$ performs a chain-of-thought audit, producing two raw lists of unimplemented and unintended changes. An LLM $\Phi_{\text{L}}$ aggregates these lists across the GDPO group into deduplicated rubrics, and $\Phi_{\text{V}}$ re-evaluates each candidate against the shared rubrics. The resulting two scalar rewards $(r^{(1)}_{i},r^{(2)}_{i})$ are each group-normalized into per-channel advantages $(A^{(1)}_{i},A^{(2)}_{i})$ , which are then combined into the final advantage that drives the flow-based GDPO update.

	Avg			ES per benchmark
Method	ES $\uparrow$	HPS $\uparrow$	IR $\uparrow$	MinEval	MagicBr.	AnyBench	Emu-Edit
FLUX.1 Kontext-dev (Base, 12B, 2025)	5.21 ±0.34	0.215 ±0.007	$-$ 0.67 ±0.14	4.72	5.25	5.46	5.40
InstructPix2Pix ( $\sim$ 1B, 2022)	2.43 ±0.34	0.211 ±0.009	$-$ 0.89 ±0.17	1.93	2.50	2.63	2.65
MagicBrush model ( $\sim$ 1B, 2023)	3.53 ±0.63	0.207 ±0.006	$-$ 0.90 ±0.07	3.25	4.45	3.04	3.38
OmniGen2 ( $\sim$ 7B, 2025)	5.08 ±0.30	0.214 ±0.006	$-$ 0.72 ±0.10	5.15	5.03	4.71	5.44
UniWorld-v1 ( $\sim$ 20B, 2025)	3.43 ±0.26	0.202 ±0.004	$-$ 0.92 ±0.05	3.38	3.40	3.78	3.15

Table 1: Final-checkpoint results on four image-editing benchmarks. We report EditScore (ES) Overall per benchmark, plus the unweighted mean across benchmarks for all three metrics ( Avg , mean

{}_{\pm\text{std}}

across the four benchmarks). The four benchmarks differ in size by

{\sim}6\times

, so we use unweighted averaging to give equal voice to each benchmark; per-benchmark HPSv2 and ImageReward are in App. C.5 . Higher is better in all columns; ES

\in[0,10]

, HPS

\in[0,1]

, IR is real-valued. Best per column in bold .

Method	Edit type
Method	act.	add	app.	bg.	color	count	impl.	mat.	move	outp.	rel.	rem.	repl.	resize	rot.	style	text	tone
Base	5.76 ∗	7.88	5.74	6.29	5.62	4.51 ∗	5.09 ∗	4.18	2.51	4.71 ∗	3.80	5.85 ∗	5.87	3.98 ∗	7.10	5.37 ∗	8.34	5.74
Arro	6.69	8.33 ∗	6.45 ∗	6.77	6.25	5.03	6.39	4.57 ∗	4.09	4.89	4.60	5.62	6.41 ∗	5.00	7.47 ∗	5.92	8.25 ∗	6.53 ∗
IP2P	1.48	4.62	3.02	2.89	4.99	0.42	2.72	2.61	1.95	2.48	2.35	1.46	2.50	2.19	3.02	3.42	0.08	5.06
MagicBr.	2.39	6.98	4.78	3.06	3.78	1.79	3.77	2.15	3.29	1.20	3.31	3.45	3.61	1.55	2.43	2.22	0.86	4.18
OmniGen2	2.56	7.18	5.93	6.72 ∗	6.46	2.32	4.42	3.98	2.16	4.62	4.02 ∗	6.30	6.74	1.61	2.70	4.55	6.78	5.71

Table 2: Per-edit-type EditScore Overall on AnyBench (50 prompts per edit type). Edit-type names follow AnyBench’s taxonomy and may differ in surface form from the six training edit types in App. E.1 (e.g., appearance_alter vs. appearance_change , action_chg. vs. action_change ). Edit types are columns and methods are rows. Higher is better. The best score in each column is shown in bold , and the second-best score is marked with ∗ . Full results are in App. C.6 .

		Avg			MinEval			MagicBrush			AnyBench			Emu-Edit
Variant	Reward	ES $\uparrow$	HPS $\uparrow$	IR $\uparrow$	ES $\uparrow$	HPS $\uparrow$	IR $\uparrow$	ES $\uparrow$	HPS $\uparrow$	IR $\uparrow$	ES $\uparrow$	HPS $\uparrow$	IR $\uparrow$	ES $\uparrow$	HPS $\uparrow$	IR $\uparrow$
(1) Base (no RL)	$\varnothing$	5.21 ±0.34	0.215 ±0.007	$-$ 0.67 ±0.14	4.72	0.212	$-$ 0.80	5.25	0.224	$-$ 0.51	5.46	0.215	$-$ 0.58	5.40	0.208	$-$ 0.78
(2) Merged scalar	$w_{1}r^{(1)}+w_{2}r^{(2)}$	4.35 ±0.37	0.211 ±0.006	$-$ 0.75 ±0.10	3.89	0.209	$-$ 0.82	4.23	0.219	$-$ 0.62	4.75	0.212	$-$ 0.72	4.52	0.205	$-$ 0.85
(3) Efficacy only	$r^{(1)}$	4.09 ±0.45	0.203 ±0.007	$-$ 0.87 ±0.17	3.46	0.197	$-$ 1.05	4.09	0.212	$-$ 0.65	4.49	0.204	$-$ 0.83	4.33	0.197	$-$ 0.94
(4) Locality only [reward-hacked]	$r^{(2)}$	ES $\approx$ 0.00 HPS $\approx$ 0.011 IR $\approx-2.25$ across all four benchmarks (degenerate noise output)
(5) Arro (full, ours)	$(r^{(1)},r^{(2)})$ + per-channel adv.	5.88 ±0.15	0.222 ±0.008	$-$ 0.51 ±0.16	5.73	0.220	$-$ 0.63	5.80	0.233	$-$ 0.29	6.07	0.221	$-$ 0.49	5.91	0.214	$-$ 0.62

Table 3: Ablation variants and results. The Reward column gives the reward signal used for RL fine-tuning; variants (2)–(4) toggle one design factor relative to the full method (5). Avg is the unweighted mean across the four benchmarks. Final-checkpoint EditScore Overall (

\uparrow

), HPSv2 (

\uparrow

), ImageReward (

\uparrow

); best per column in bold . Variant (4) reward-hacks to a degenerate noise output and is collapsed to a single line (see text).

Comparison	Win	Tie	Loss
ARRO vs. EditScore(PF, Cons)-GDPO	366/600 (61.0%)	94/600 (15.7%)	140/600 (23.3%)
OmniGen2 + ARRO vs. OmniGen2	321/600 (53.5%)	107/600 (17.8%)	172/600 (28.7%)

Table 4: Blinded pairwise judgments on 60 MinEval examples by ten evaluators (600 judgments per comparison). Win, tie, and loss are from the ARRO variant’s perspective; judgments share 60 underlying image pairs.

Benchmark	OmniGen2	OmniGen2 + ARRO	ES gain
MagicBrush	5.03	5.58	+0.55
AnyBench	4.71	5.22	+0.51
Emu-Edit	5.44	5.99	+0.55
Average	5.06	5.59	+0.53

Table 5: OmniGen2 transfer on three unseen benchmarks. EditScore (ES) is higher-is-better; the average is unweighted. Full HPSv2 and ImageReward results are in App. C.3 .

Appendix figures & tables13 assets

Supplementary material from the paper’s appendix.

Appendix

Method	OffTarget-L1 $\downarrow$	Improvement vs. Base
FLUX.1 Kontext-dev (Base)	0.20366	—
EditScore-Top1-SFT	0.19952	2.0%
EditScore(PF, Cons)-GDPO	0.18975	6.8%
ARRO	0.18659	8.4%

Appendix

Table 6: Off-target pixel change on 600 randomly selected evaluation examples. Relative improvement is computed against the base editor. Lower OffTarget-L1 is better.

Method	PF $\uparrow$	Cons. $\uparrow$	PQ $\uparrow$	ES $\uparrow$	HPSv2 $\uparrow$	IR $\uparrow$
Base	5.9920	8.1573	6.3393	4.7237	0.2120	$-0.7991$
Raw audit counts	5.0843	7.0832	4.0630	3.2779	0.1911	$-1.0830$
EditScore Top-10 SFT	5.7700	8.2827	6.2713	4.5529	0.2091	$-0.8467$
ARRO Top-10 SFT	5.7893	8.3240	6.3933	4.6289	0.2082	$-0.8782$
ARRO-GDPO	7.1113	8.1700	6.5693	5.6690	0.2196	$-0.6309$

Appendix

Table 7: Reward-pipeline and Top-10 SFT controls on the same 600-example MinEval evaluation. All methods use the same EditScore, HPSv2, and ImageReward protocols.

Reward	MinEval	MagicBrush	AnyBench	Emu-Edit	Average
EditScore(PF, Cons)	3.566	3.233	4.379	3.932	3.777
ARRO verified issue counts	5.730	5.800	6.070	5.910	5.880

Appendix

Table 8: EditScore overall under fixed GDPO training with different reward signals. Average is unweighted across four benchmarks.

Requested modifications	5	10	15	20	25	30	35	40	45	50
ARRO unintended count	0	0	0	0	0	0	0	0	0	0
EditScore Consistency	9.200	8.333	8.867	8.467	8.733	8.667	8.533	9.240	8.867	9.000

Appendix

Table 9: CLEVR diagnostic: every specified modification is implemented and the ground-truth unintended-change count is zero. Each column reports a five-sample mean.

VLM + LLM	GT=0	GT=1	GT=2	GT=3	MAE $\downarrow$	Exact acc. $\uparrow$
Qwen3 4B + 4B	0.80	1.40	1.80	3.40	0.75	0.40
Qwen3 8B + 8B	0.80	1.00	2.00	2.80	0.55	0.50
Qwen3 32B + 30B	0.40	1.20	1.80	3.00	0.30	0.70

Appendix

Table 10: Evaluator-size diagnostic on 20 CLEVR outputs. GT columns show the average predicted unintended-change count among five examples with that ground-truth count.

Check	No error confirmed by both	Error confirmed by both
Repeated issues merged correctly	560/565 (99.1%)	5/565 (0.9%)
Final issue visually supported	553/565 (97.9%)	12/565 (2.1%)
No evident issue omitted	58/60 (96.7%)	2/60 (3.3%)

Appendix

Table 11: Expert validation of the current reward pipeline. Fractions use issues for the first two rows and candidate groups for the third.

Benchmark	Model	ES $\uparrow$	HPSv2 $\uparrow$	ImageReward $\uparrow$
MagicBrush	OmniGen2	5.03	0.221	$-0.59$
	OmniGen2 + ARRO	5.58	0.228	$-0.33$
AnyBench	OmniGen2	4.71	0.215	$-0.68$
	OmniGen2 + ARRO	5.22	0.221	$-0.58$
Emu-Edit	OmniGen2	5.44	0.207	$-0.77$
	OmniGen2 + ARRO	5.99	0.213	$-0.60$

Appendix

Table 12: OmniGen2 transfer on three unseen benchmarks. Average is the unweighted mean of benchmark scores.

	MinEval			MagicBrush			AnyBench			Emu-Edit			Avg
Method	PF	Cons	PQ	PF	Cons	PQ	PF	Cons	PQ	PF	Cons	PQ	PF	Cons	PQ
FLUX.1 Kontext-dev (Base)	6.00	8.12	6.39	6.74	8.50	6.57	6.64	8.48	6.70	7.03	8.33	6.15	6.60	8.36	6.45
InstructPix2Pix	2.31	6.13	6.24	3.24	6.65	6.20	3.34	6.98	6.39	3.34	6.60	6.06	3.06	6.59	6.22
MagicBrush model	4.05	5.94	4.82	5.63	7.22	5.29	4.44	5.63	4.37	4.73	6.13	5.02	4.71	6.23	4.87
OmniGen2	6.03	8.77	7.13	6.11	9.04	7.22	6.03	8.46	6.76	6.78	8.54	6.49	6.24	8.70	6.90
UniWorld-v1	5.08	5.56	3.95	5.12	6.18	3.97	5.27	5.97	4.67	5.14	5.64	3.80	5.15	5.84	4.10

Appendix

Table 13: Per-benchmark EditScore sub-scores (PF, Cons, PQ) for all methods, complementing the Overall numbers in Table 1 . All sub-scores lie in

[0,10]

\uparrow

. Avg is the unweighted mean across the four benchmarks. Best per column in bold .

	MinEval		MagicBrush		AnyBench		Emu-Edit
Method	HPS	IR	HPS	IR	HPS	IR	HPS	IR
FLUX.1 Kontext-dev (Base)	0.212	$-$ 0.80	0.224	$-$ 0.51	0.215	$-$ 0.58	0.208	$-$ 0.78
InstructPix2Pix	0.220	$-$ 0.67	0.213	$-$ 0.96	0.212	$-$ 0.85	0.199	$-$ 1.08
MagicBrush model	0.211	$-$ 0.84	0.211	$-$ 0.86	0.205	$-$ 0.88	0.199	$-$ 1.00
OmniGen2	0.211	$-$ 0.83	0.221	$-$ 0.59	0.215	$-$ 0.68	0.207	$-$ 0.77
UniWorld-v1	0.203	$-$ 0.92	0.206	$-$ 0.90	0.204	$-$ 0.87	0.196	$-$ 0.99

Appendix

Table 14: Per-benchmark HPSv2 (HPS) and ImageReward (IR) for all methods. Best per column in bold . EditScore Overall per benchmark is in Table 1 .

Method	Edit type
Method	act.	add	app.	bg.	color	count	impl.	mat.	move	outp.	rel.	rem.	repl.	resize	rot.	style	text	tone
Base	0.266 ∗	0.266 ∗	0.245 ∗	0.210	0.223	0.177	0.240	0.221	0.210	0.191 ∗	0.215	0.169	0.234 ∗	0.221	0.183 ∗	0.191	0.233	0.177
Arro	0.269	0.268	0.255	0.218	0.234	0.182	0.251	0.227	0.219 ∗	0.193	0.219	0.165	0.241	0.226 ∗	0.184	0.202	0.231	0.187
IP2P	0.247	0.229	0.222	0.198	0.230	0.203	0.226	0.230 ∗	0.221	0.170	0.218 ∗	0.207	0.222	0.230	0.173	0.190	0.227	0.166
MagicBr.	0.233	0.240	0.228	0.222 ∗	0.221	0.181	0.216	0.217	0.211	0.165	0.203	0.193 ∗	0.217	0.216	0.161	0.175	0.208	0.190 ∗
OmniGen2	0.261	0.252	0.242	0.222	0.235 ∗	0.172	0.247 ∗	0.234	0.187	0.188	0.205	0.173	0.229	0.217	0.176	0.200	0.236	0.190

Appendix

Table 15: Per-edit-type HPSv2 on AnyBench (50 prompts per edit type), complementing Table 2 in the main paper. Edit-type columns and method order match Table 2 . Higher is better. The best score in each column is shown in bold , and the second-best score is marked with ∗ .

Method	Edit type
Method	act.	add	app.	bg.	color	count	impl.	mat.	move	outp.	rel.	rem.	repl.	resize	rot.	style	text	tone
Base	0.42 ∗	0.79	0.27 ∗	$-$ 0.81	$-$ 0.15	$-$ 1.33	0.05	$-$ 0.79	$-$ 0.46	$-$ 1.12	$-$ 0.49	$-$ 1.76	$-$ 0.66	$-$ 0.44	$-$ 1.57 ∗	$-$ 1.15	$-$ 0.09	$-$ 1.18
Arro	0.56	0.78 ∗	0.42	$-$ 0.59	0.08 ∗	$-$ 1.21	0.32	$-$ 0.54 ∗	$-$ 0.35	$-$ 0.99 ∗	$-$ 0.40	$-$ 1.80	$-$ 0.76 ∗	$-$ 0.47 ∗	$-$ 1.55	$-$ 1.02	$-$ 0.21	$-$ 1.07
IP2P	$-$ 0.43	$-$ 0.61	$-$ 0.86	$-$ 1.16	$-$ 0.50	$-$ 0.81	$-$ 0.53	$-$ 0.74	$-$ 0.38 ∗	$-$ 1.44	$-$ 0.44 ∗	$-$ 1.10	$-$ 1.18	$-$ 0.55	$-$ 1.73	$-$ 1.19	0.02	$-$ 1.61
MagicBr.	$-$ 0.36	$-$ 0.14	$-$ 0.49	$-$ 0.58 ∗	$-$ 0.60	$-$ 1.22	$-$ 0.55	$-$ 1.01	$-$ 0.67	$-$ 1.43	$-$ 0.63	$-$ 1.34 ∗	$-$ 1.12	$-$ 0.81	$-$ 1.88	$-$ 1.27	$-$ 0.62	$-$ 1.15 ∗
OmniGen2	0.05	0.34	0.02	$-$ 0.54	$-$ 0.02	$-$ 1.48	$-$ 0.05	$-$ 0.51	$-$ 1.05	$-$ 0.99	$-$ 0.61	$-$ 1.91	$-$ 0.94	$-$ 0.59	$-$ 1.70	$-$ 1.06 ∗	$-$ 0.07 ∗	$-$ 1.15 ∗

Appendix

Table 16: Per-edit-type ImageReward on AnyBench (50 prompts per edit type), complementing Table 2 in the main paper. Edit-type columns and method order match Table 2 . Higher is better. The best score in each column is shown in bold , and the second-best score is marked with ∗ .

Optimizer
Parameter	Setting
Optimizer	AdamW
Learning rate	$1\!\times\!10^{-5}$
Weight decay	$1\!\times\!10^{-4}$
Gradient-norm clip	$1.0$
PPO ratio clip	$5\!\times\!10^{-3}$

Appendix

Table 17: Key hyperparameters of Arro fine-tuning on FLUX.1 Kontext-dev.

Reward
Parameter	Setting
Channel-1 weight $w_{1}$	$0.3$
Channel-2 weight $w_{2}$	$0.7$
Auditor VLM	Qwen3-VL-30B-A3B-Instruct [ Qwen Team, 2025 ]
Aggregator LLM	Qwen3-30B-A3B-Instruct-2507 [ Qwen Team, 2025 ]
Compute

Appendix

Table 18: Reward, compute, and inference settings for FLUX.1 Kontext-dev experiments.

Explore similar work

Leveraging Verifier-Based Reinforcement Learning in Image Editing
Apr 30, 2026Hanzhong Guo, Jie Wu, Jie Liu +6Image Editing Contrastive Language-Image Pre-Training Model
School of Computing and Data Science, The University of Hong Kong · ByteDance Seed · CUHK +1
Refinement Is Inherently Editable: Training-Free Prompt-to-Prompt Image Editing with Generative Refinement Network
Sep 17, 2026Yulong Chen, Ziqian Zhang, Haoyu Zhang +4Diffusion-Based Image Editing Image Editing
City University of Hong Kong (Dongguan), Guangdong, China · City University of Hong Kong, Hong Kong, China · Mohamed bin Zayed University of Artificial Intelligence, Masdar, Abu Dhabi
RewardHarness: Self-Evolving Agentic Post-Training
May 9, 2026Yuxuan Zhang, Penghui Du, Bo Li +11Progress Reward Modeling Multi-Image Editing Benchmarks
University of British Columbia · Kolors Team, Kuaishou Technology · University of Waterloo +4

Apr 30, 2026, cs.CVJ/K move · Enter open · S save

Leveraging Verifier-Based Reinforcement Learning in Image Editing

Hanzhong Guo, Jie Wu, Jie Liu, Yu Gao+5

School of Computing and Data Science, The University of Hong Kong · ByteDance Seed · CUHK+1

Image Editing Contrastive Language-Image Pre-Training Model Reinforcement Learning With Verifiable Reward

Open paper AskPDF arXiv

While Reinforcement Learning from Human Feedback (RLHF) has become a pivotal paradigm for text-to-image generation, its application to image editing remains largely unexplored. A key bottleneck is the lack of a robust general reward model for all editing tasks. Existing edit reward models usually give overall scores without detailed checks, ignoring different instruction requirements and causing biased rewards. To address this, we argue that the key is to move from a simple scorer to a reasoning verifier. We introduce Edit-R1, a framework that builds a chain-of-thought (CoT) verifier-based reasoning reward model (RRM) and then leverages it for downstream image editing. The Edit-RRM breaks instructions into distinct principles, evaluates the edited image against each principle, and aggregates these checks into an interpretable, fine-grained reward. To build such an RRM, we first apply supervised fine-tuning (SFT) as a ``cold-start'' to generate CoT reward trajectories. Then, we introduce Group Contrastive Preference Optimization (GCPO), a reinforcement learning algorithm that leverages human pairwise preference data to reinforce our pointwise RRM. After building the RRM, we use GRPO to train editing models with this non-differentiable yet powerful reward model. Extensive experiments demonstrate that our Edit-RRM surpasses powerful VLMs such as Seed-1.5-VL and Seed-1.6-VL as an editing-specific reward model, and we observe a clear scaling trend, with performance consistently improving from 3B to 7B parameters. Moreover, Edit-R1 delivers gains to editing models like FLUX.1-kontext, highlighting its effectiveness in enhancing image editing.

Abstract

Explore similar work

Leveraging Verifier-Based Reinforcement Learning in Image Editing

Refinement Is Inherently Editable: Training-Free Prompt-to-Prompt Image Editing with Generative Refinement Network

RewardHarness: Self-Evolving Agentic Post-Training