Paper 2512.17504
InsertAnywhere: Geometrically Grounded and Optics-Aware Video Object Insertion
- Published
- Dec 2025
- Research lab
- Independent
- Citations
- 0
- GitHub
- 89 stars
01 In brief
Summary
InsertAnywhere is a framework for video object insertion (VOI) that addresses limitations in 4D scene understanding and optical effects.
It uses a two-stage pipeline: first, a 4D-aware mask generation module reconstructs the video into a 4D scene, allowing users to anchor an object's 3D pose in one frame, then propagates it via scene flow tracking to produce temporally coherent masks that handle occlusions and moving surfaces.
Second, an optics-aware video synthesis stage uses a diffusion model (Wan2.1-VACE-14B) fine-tuned with LoRA on a new dataset, ROSE++, which provides quadruplets of source video, target video, mask, and reference image.
A novel Optics-Aware Representation Alignment loss aligns features from a fine mask with those from an extended mask (including shadows/reflections), enabling the model to synthesize optical effects beyond the object boundary.
First-frame anchoring with an image insertion model improves fidelity.
Experiments on a new benchmark, VOIBench (200 clips), show InsertAnywhere outperforms commercial tools (Kling, Pika-Pro) and open-source methods (AnyV2V, ReVideo, Señorita, VACE) in subject consistency (CLIP-I 0.8132, DINO-I 0.5669) and VBench metrics.
Ablations confirm each component contributes, and the alignment loss improves optics-region PSNR by +7.3 dB.
02 From the paper
Abstract
Recent advances in diffusion models have enabled impressive video editing capabilities, yet production-grade Video Object Insertion (VOI) remains challenging due to inadequate 4D scene understanding and a lack of proper optical interactions, such as shadows and reflections. To address these limitations, we present InsertAnywhere, a comprehensive VOI framework that achieves geometrically grounded object placement and optics-aware video synthesis. Our approach first leverages a 4D-aware mask generation module that allows users to anchor an object's 3D pose in a single frame. The framework automatically propagates this placement across the video, accurately handling local scene dynamics and occlusions. To synthesize realistic physical lighting interactions, we introduce Optics-Aware Representation Alignment, a novel strategy that utilizes an extended mask to guide feature extraction, enabling optical effects to seamlessly extend beyond the inserted object's boundary. Finally, to overcome the lack of training data for such phenomena, we construct and open-source ROSE++, a specialized quadruplet dataset tailored for the supervised learning of optical effects. Extensive experiments demonstrate that InsertAnywhere produces geometrically plausible and photometrically realistic insertions in complex real-world scenarios, significantly outperforming existing research and commercial generative tools.