In this paper, we tackle the problem of performing consistent and unified modifications across a set of related images. This task is particularly challenging because these images may vary significantly in pose, viewpoint, and spatial layout. Achieving coherent edits requires establishing reliable correspondences across the images, so that modifications can be applied accurately to semantically aligned regions. To address this, we propose GroupEditing, a novel framework that builds both explicit and implicit relationships among images within a group. On the explicit side, we extract geometric correspondences using VGGT, which provides spatial alignment based on visual features. On the implicit side, we reformulate the image group as a pseudo-video and leverage the temporal coherence priors learned by pre-trained video models to capture latent relationships. To effectively fuse these two types of correspondences, we inject the explicit geometric cues from VGGT into the video model through a novel fusion mechanism. To support large-scale training, we construct GroupEditData, a new dataset containing high-quality masks and detailed captions for numerous image groups. Furthermore, to ensure identity preservation during editing, we introduce an alignment-enhanced RoPE module, which improves the model's ability to maintain consistent appearance across multiple images. Finally, we present GroupEditBench, a dedicated benchmark designed to evaluate the effectiveness of group-level image editing. Extensive experiments demonstrate that GroupEditing significantly outperforms existing methods in terms of visual quality, cross-view consistency, and semantic alignment.
Figures & tables
Figure 1: Data curation pipeline. Our pipeline processes human-written instructions through text-to-image generation, followed by quality evaluation (consistency and aesthetic assessment) and annotation generation (object segmentation and text description). The pipeline automatically produces high-quality training pairs with precise masks and detailed captions.
Figure 2: Overview of proposed method . Given a series of images and their corresponding masks, we propose a novel framework for editing while ensuring the consistency of multiple images. To achieve fine-grained spatial alignment, we introduce Geometry-enhanced RoPE (GE-RoPE), which enhances the model’s ability to maintain consistent spatial relationships across different frames, and Identity-RoPE for better consistent identity preservation.
Figure 3: Visual illustration about correspondence using VGGT. Comparisons of the input and our correspondence prediction using VGGT. The results without VGGT show less accurate alignment, while the inclusion of VGGT improves consistency and precision in the correspondence.
Algorithm 1 Identity-RoPE for pixel alignment
Figure 4: 3D reconstruction using editing results. The top row shows input images, the middle row displays the edited results, and the bottom row demonstrates 3D reconstruction using Must3R, aligning 2D points to create a 3D model of the modified bus.
Figure 5: Image customization using editing results. Customized generation is achieved by embedding the edited concepts into the generative model, leveraging the outputs from our consistent editing method.
Figure 6: Visual comparison with SOTA image editing on local and global editing .The top row shows results for different methods on local edits. The bottom row presents our method alongside other techniques for global editing. Our approach demonstrates superior consistency and high-quality edits across both local and global changes. The text prompts are “ A cartoon fox with futuristic robotic armor and orange details. ” and “ A jeep vehicle in outdoor settings, including snow-covered style, rendered in soft watercolor painting style. ” .
Figure 7: Visual ablation about proposed modules. We provide a visual ablation study on the effectiveness of various mechanisms. The prompt is “ The penguin has a black and white body with orange feet and beak. ”
Method
Quantitative metrics
User study
CLIP-Score↑
Aesthetic-Score↑
DINO-Score↑
Editing Consis. ↑
PSNR↑
Iden. Consis.↓
Aesthetic.↓
App. Fidelity↓
Overall↓
Anydoor [ 11 ]
0.2728
4.72
0.7208
0.8697
0.6182
3.56
3.23
3.60
3.32
OminiControl [ 57 ]
0.2902
5.10
0.7326
0.8676
0.6457
2.84
2.84
1.96
2.66
Edicho [ 3 ]
0.3059
4.89
0.8080
0.8988
0.6935
1.93
2.47
2.93
1.95
Ours
0.3122
5.39
0.8168
0.9239
0.7624
1.67
1.46
1.50
1.47
Table 1: Comparison with state-of-the-art image editing on local editing . Red and Blue denote the best and second best results.
Method
Quantitative metrics
CLIP-Score↑
Aesthetic-Score↑
DINO-Score↑
Editing Consis.↑
StyleAligned [ 21 ]
0.2910
4.75
0.8645
0.8714
Edicho [ 3 ]
0.2920
4.59
0.8714
0.8859
OminiControl [ 57 ]
0.2918
4.94
0.8553
0.8965
Ours
0.2987
5.48
0.9287
0.9147
Table 2: Comparison with state-of-the-art image editing on global editing . Red and Blue denote the best and second best results, respectively.
Method
Quantitative metrics
CLIP-Score↑
Aesthetic-Score↑
DINO-Score↑
Editing Consis.↑
W/o VGGT [ 21 ]
0.2728
4.72
0.7208
0.8616
W/o Ge-RoPE [ 3 ]
0.2902
4.89
0.7326
0.8697
W/o Identity-RoPE [ 57 ]
0.2902
4.89
0.7326
0.9108
Ours
0.3122
5.39
0.8168
0.9239
Table 3: Quantitative ablation . Red and Blue denote the best and second best results, respectively.
In this paper, we tackle the problem of performing consistent, unified modifications to a multi-shot video sequence. This task is particularly challenging because multi-shot videos consist of discontinuous temporal segments that vary significantly in viewpoint, camera scale, and subject pose, leading to severe identity drift and cumulative error propagation. Achieving coherent edits requires establishing reliable cross-shot semantic awareness to maintain stable subject appearance and visual continuity across these disjointed boundaries. To address this, we propose MSEditor, the first framework designed specifically for consistent multi-shot video editing. To overcome the scarcity of high-quality multi-shot training data, we repurpose existing multi-view video datasets to provide robust cross-shot supervision. Architecturally, we introduce a Supervisory Adapter that injects this cross-shot information into the diffusion backbone, enabling the model to learn identity-consistent representations. Furthermore, to effectively mitigate cumulative errors and ensure long-range temporal coherence, we design a Cross-Shot Packing strategy that dynamically aggregates information from semantically related shots within the self-attention window. Extensive experiments demonstrate that MSEditor significantly outperforms existing methods on our curated multi-shot video editing benchmark in terms of identity preservation, temporal stability, and overall visual quality.
Instruction-guided image editing has seen remarkable progress with models like FLUX.2 and Qwen-Image-Edit, yet they still struggle with complex scenarios with multiple similar instances each requiring individual edits. We observe that state-of-the-art models suffer from severe over-editing and spatial misalignment when faced with multiple identical instances and composite instructions. To this end, we introduce a comprehensive benchmark specifically designed to evaluate fine-grained consistency in multi-instance and multi-instruction settings. To address the failures of existing methods observed in our benchmark, we propose Multi-Instance Regional Alignment via Guided Editing (MIRAGE), a training-free framework that enables precise, localized editing. By leveraging a vision-language model to parse complex instructions into regional subsets, MIRAGE employs a multi-branch parallel denoising strategy. This approach injects latent representations of target regions into the global representation space while maintaining background integrity through a reference trajectory. Extensive evaluations on MIRA-Bench and RefEdit-Bench demonstrate that our framework significantly outperforms existing methods in achieving precise instance-level modifications while preserving background consistency. Our benchmark and code are available at https://github.com/ZiqianLiu666/MIRAGE.
Ziqian Liu, Stephan Alaniz
LTCI, Télécom Paris, Institut Polytechnique de Paris, France
Recent advancements in text-guided 3D editing, particularly with 3D Gaussian Splatting (3DGS), have shown promise but still lag behind the flexibility of 2D editing tools. Existing methods face a critical trade-off between preserving multi-view consistency and allowing structural modifications. Image-based editors can also struggle to apply the intended edit to objects viewed from unusual angles. To address these limitations, we introduce VidEdit3D, a framework that reformulates 3D scene editing as a trajectory-based video editing task. By drawing a parallel between camera movement in static 3D scenes and temporal dynamics in video, we leverage the robust spatiotemporal priors of pre-trained Video Diffusion Models (VDMs) to encourage semantic and structural coherence across views. Specifically, we propose Trajectory-Guided Rendering (TGR) to select informative keyframes and construct smooth video clips aligned with VDM input requirements. To propagate edits from an anchor view to these clips, we develop Flow-Aligned Video Editing (FAVE), a training-free framework incorporating an Editing Consistency Guidance (ECG) mechanism. ECG contrasts source and edited image conditions at a shared latent state, using scheduled source-image amplification to strengthen adherence to the edited reference while maintaining original scene context. Finally, we introduce Region-Selective 3DGS Refinement (RS-GS) to strictly confine geometric and appearance updates to target primitives, reducing background attribute leakage. Experiments demonstrate stronger instruction alignment and better multi-view consistency than the baselines.
Yanming Yang, Chenxi Song, Tong Zhao +5
Westlake University · University of Leeds · Nanyang Technological University +1