arXiv:2605.30045v1 Announce Type: new 
Abstract: Video object removal frequently struggles to simultaneously eliminate target objects and their associated physical effects (e.g., smoke, reflections, light, and ripples) in out-of-domain scenarios due to complex spatiotemporal ambiguities. While existing methods primarily rely on spatial masks, they often fail to capture weakly correlated effects, and the potential of explicit textual guidance remains underexplored. Furthermore, a fundamental optimization conflict exists in removal models between high-level semantic generalization and precise pixel-level background preservation. To address these challenges, we propose GenEraser, a novel framework for generalized and high-fidelity video object and effect removal. First, we introduce a Multi-Conditional Mixture-of-Experts (MC-MoE) paired with Bipartite Text guidance to fully exploit the multimodal priors of Diffusion Transformers, significantly enhancing the identification of complex effects. Second, a Learnable Deep ``CFG'' Fusion mechanism (LD-CFG) is developed to adaptively balance the relative dominance of mask and textual conditions across diverse scenarios. Finally, we propose a Decoupled Expert Architecture, comprising a Locator and a Preserver, to mitigate the inherent trade-off between semantic generalization and pixel alignment. Extensive experiments demonstrate that our GenEraser surpasses recent state-of-the-art approaches, achieving significant quantitative improvements (e.g., $2.16$ dB and $1.44$ dB on the ROSE Benchmark and VOR-Eval, respectively) while maintaining exceptionally robust generalization in open-world scenarios. https://cyqii.github.io/GenEraser.github.io/

تم تقديم إطار عمل جديد يسمى GenEraser لإزالة الكائنات من الفيديو، حيث يتناول التحديات المتعلقة بإزالة الكائنات المستهدفة وآثارها الفيزيائية في السيناريوهات المعقدة. يستفيد هذا الإطار من مزيج من الخبراء متعدد الشروط وتوجيه نص ثنائي لتحسين تحديد التأثيرات المعقدة، متجاوزًا قيود الطرق الحالية التي تعتمد بشكل أساسي على الأقنعة المكانية.

Se ha introducido un nuevo marco llamado GenEraser para la eliminación de objetos en video, abordando los desafíos de eliminar objetos objetivo y sus efectos físicos en escenarios complejos. Este marco aprovecha el Multi-Conditional Mixture-of-Experts y la orientación textual bipartita para mejorar la identificación de efectos intrincados, superando las limitaciones de los métodos existentes que se basan principalmente en máscaras espaciales.

Un nouveau cadre nommé GenEraser a été introduit pour la suppression d'objets vidéo, abordant les défis d'élimination des objets cibles et de leurs effets physiques dans des scénarios complexes. Ce cadre exploite le Multi-Conditional Mixture-of-Experts et l'orientation textuelle bipartite pour améliorer l'identification des effets complexes, surmontant les limitations des méthodes existantes qui s'appuient principalement sur des masques spatiaux.

A new framework named GenEraser has been introduced for video object removal, addressing the challenges of eliminating target objects and their physical effects in complex scenarios. This framework leverages Multi-Conditional Mixture-of-Experts and Bipartite Text guidance to enhance the identification of intricate effects, overcoming limitations of existing methods that primarily rely on spatial masks.