Diffusion Models are Secretly Zero-Shot 3DGS Harmonizers
Vsevolod Skorokhodov [email protected], Nikita Durasov [email protected], Pascal Fua [email protected] · 2025
Read Diffusion Models are Secretly Zero-Shot 3DGS Harmonizers by Vsevolod Skorokhodov [email protected], Nikita Durasov [email protected], Pascal Fua [email protected] (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-06740.txt.
Description
Vsevolod Skorokhodov, Nikita Durasov, and Pascal Fua found in 2025 that image diffusion models quietly know how lighting should look, and used that knowledge to seat objects into three-dimensional scenes. Gaussian splatting renders scenes as clouds of soft blobs, and dropping a new object in is easy; making it belong, with the scene's light and shadows on it, is the unsolved part.
Their method inserts the object, then optimizes its blob parameters against an objective drawn from denoising score methods, so a diffusion model grades the blend. A personalization trick keeps the object's shape and texture intact through the relighting. The catalog creator field appends email addresses to the names; the head byline governs. Readers get object insertion that respects the room's light, judged by a model that has seen millions of rooms.
Catalog record
| CREATOR | Vsevolod Skorokhodov [email protected], Nikita Durasov [email protected], Pascal Fua [email protected] |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 75 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-06740