# Diffusion Model Enables Lighting-Consistent Object Transfer Between 3D Gaussian Splatting Scenes

_Research · published 2026-06-23_

Researchers from Inria, Google DeepMind, and Eyeline Labs have introduced a diffusion-based method for transferring objects between 3D Gaussian Splatting (3DGS) scenes while harmonizing their lighting conditions — a longstanding challenge in VFX compositing. When an object is extracted from one captured scene and naively pasted into another, mismatched lighting makes the result look artificial; the new method resolves this by training a diffusion model on heterogeneous image pairs of inconsistent and consistent composites, then consolidating the harmonized views back into a 3DGS representation via a post-optimization step.

The full pipeline — covering extraction, composite rendering, per-view harmonization, and 3DGS reconstruction — is presented at the 2026 Eurographics Symposium on Rendering (Computer Graphics Forum). The authors report visually compelling improvements over prior methods, with potential applications spanning VFX, architecture, interior design, and marketing. Code and models are publicly available on GitHub.

## Sources
- [repo-sam.inria.fr](https://repo-sam.inria.fr/nerphys/dot3d/)
---
Canonical: https://genbuzz.news/posts/diffusion-model-enables-lighting-consistent-object-transfer-between-3d-gaussian-splatting-scenes