Smart-Insertion-V: Photorealistic Video Insertion via a Closed-Loop Feedback Dual-Stream Framework

2026-05-22

AI Summary

This work addresses the challenge of unnatural blending in mask-free video object insertion caused by domain discrepancies between the reference object and the source scene’s visual style. To this end, we propose an end-to-end dual-stream framework that jointly performs video insertion and image style transfer, enhanced with a closed-loop feedback mechanism to improve robustness. Our approach introduces a novel Dual-World-View RoPE positional encoding to disambiguate multi-source signals and incorporates a decoupled guidance module that fuses semantic reasoning from vision-language models with temporal guidance cues. Additionally, we construct the first open-source dataset dedicated to this task, filling a critical gap in the field. Experiments demonstrate that our method achieves state-of-the-art performance, significantly enhancing stylistic harmony while maintaining accurate spatial placement.

Abstract

Mask-free video object insertion has emerged as a challenging task, requiring harmonious integration of reference objects into source videos. However, existing methods struggle when references exhibit severe stylistic domain gaps with the source scene. To overcome this, we propose \textit{\textbf{Smart-Insertion-V}}, an end-to-end \textbf{Dual-Stream} framework that concurrently conducts video insertion and image style transfer. Within this framework, the image stream synchronously guides the video generation process, while a \textbf{Closed-loop Feedback} mechanism is further incorporated to ensure robust insertion. Inevitably, integrating these diverse conditioning signals results in feature entanglement and style leakage. To tackle this issue, we design \textbf{Dual-World-View RoPE} to distinguish different signals via spatial-temporal offsets without incurring heavy training overhead. Furthermore, to facilitate spatial grounding and stylistic adaptation, we introduce a \textbf{Decoupled Guidance Module} that leverages a Vision-Language Model for semantic reasoning while preserving original temporal guidance with native text encoder. To bridge data gap for harmonious reference insertion task, we propose a data curation pipeline and will release an \textbf{open-source dataset}. Experiments demonstrate that our method can insert objects into plausible positions while achieving the most harmonious results.

Problem

• video object insertion
• photorealistic insertion
• style harmonization
• mask-free insertion
• domain gap

Innovation

• Dual-Stream Framework
• Closed-loop Feedback
• Dual-World-View RoPE
• Decoupled Guidance Module
• Mask-free Video Insertion