MirrorWorld: Taming Video Diffusion Models for Mirror Reflection Generation

2026-08-07

AI Summary

This work addresses the common failure of video diffusion models to generate physically plausible mirror reflections, often resulting in semantic inconsistencies or spatial distortions due to neglecting the geometric and semantic coherence between scenes and their reflections. To overcome this, the authors propose a novel video inpainting framework specifically designed for mirror reflection synthesis, which decouples reflection generation into two complementary subtasks: “what to reflect” (semantic content) and “how to arrange it” (spatial layout). The approach leverages Semantic Relation Distillation (SRD) to transfer semantic correlations from a frozen vision foundation model and incorporates Geometric Transformation Alignment (GTA) to model the spatial transformation inherent in reflections. Evaluated on a newly established unified benchmark for video mirror reflection reconstruction, the method significantly outperforms existing image-level reflection generation and video inpainting baselines, achieving high-fidelity and spatially consistent mirror reflections.

Abstract

Recent advances in video diffusion models (VDMs) have enabled high-fidelity video synthesis. However, generating mirror reflections remains challenging because the content within a mirror must remain consistent with the surrounding scene. Existing VDMs are not specifically designed to model scene-to-mirror relationships, which can lead to reflections with incorrect content or inconsistent spatial arrangements. We observe that mirror reflection generation involves two complementary challenges: determining what scene content should be reflected and how the reflected content should be spatially arranged within the mirror region. Motivated by this observation, we propose MirrorWorld, a reflection-aware video inpainting framework that models scene-to-mirror relationships during generation. Specifically, we introduce Semantic Relation Distillation (SRD), which transfers relational information from a frozen visual foundation model to encourage semantic associations between visible scene content and mirror regions. We further propose Geometric Transformation Alignment (GTA), which learns a transformation that guides the spatial arrangement of reflected content. The two components play complementary roles, with SRD modeling what should be reflected and GTA modeling how it should be arranged. To facilitate research on this problem, we construct a benchmark for video mirror reflection generation by repurposing four existing video mirror datasets into a unified reflection reconstruction task. Experimental results show that MirrorWorld achieves improved reflection reconstruction quality over representative image-based reflection generation methods and strong video inpainting baselines.

Problem

• mirror reflection
• video diffusion models
• scene-to-mirror consistency
• spatial arrangement
• reflection generation

Innovation

• Semantic Relation Distillation
• Geometric Transformation Alignment
• Video Diffusion Models
• Mirror Reflection Generation
• Reflection-aware Inpainting