IDAG-Edit: Multi-Object Video Editing via Instance-Decoupled Attention and Guidance

2026-06-20

AI Summary

This work addresses the limitations of diffusion models in multi-object video editing, where attention leakage, identity drift, and temporal instability often compromise both precision and temporal consistency. To overcome these challenges, the authors propose a training-free framework that enables fine-grained, object-level control through layout-guided attention modulation combined with instance-level masks. Central to their approach is an instance-decoupled attention mechanism designed to preserve the identity and temporal coherence of individual objects throughout the video sequence. Experimental results demonstrate that the proposed method significantly outperforms existing techniques in both qualitative and quantitative evaluations, effectively enhancing controllability and temporal stability in multi-object video editing scenarios.

Abstract

Diffusion-based video editing has made significant progress; however, achieving precise and temporally consistent object-level control, especially in multi-object scenarios, remains challenging due to attention leakage, identity drift, and unstable temporal dynamics. In this work, we propose IDAGEdit, a training-free framework for fine-grained multi-object video editing with strong temporal consistency. The framework adopts Layout-guided Attention Modulation to facilitate coherent multi-object editing, while Instance-level Masks are introduced to preserve individual object identity and enforce localized attention within each object region, thereby enabling fine-grained, object-level editing. Extensive qualitative and quantitative evaluations demonstrate that our method improves temporal stability and multi-object controllability over state-of-the-art video editing approaches.

Problem

• multi-object video editing
• temporal consistency
• attention leakage
• identity drift
• object-level control

Innovation

• Instance-Decoupled Attention
• Layout-guided Attention Modulation
• Instance-level Masks
• Temporal Consistency
• Training-free Video Editing