PanoDreamer: Consistent Text to 360-Degree Scene Generation

Zhexiao Xiong, Zhang Chen, Zhong Li, Nathan Jacobs, [0.5ex] OPPO US Research Center · 2025

Read PanoDreamer: Consistent Text to 360-Degree Scene Generation by Zhexiao Xiong, Zhang Chen, Zhong Li, Nathan Jacobs, [0.5ex] OPPO US Research Center (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-2504-05152.txt.

Description

Turning a text prompt into a whole 3D place you can walk around is a standing challenge for virtual reality and games. Five authors present PanoDreamer, a system that builds a full 360-degree scene from text, a reference image, or both. A language model plans the scene, and a warp-and-refine pipeline drafts the first views and stitches them into a panorama.

That panorama is lifted into a point cloud, the skeleton of the 3D world. The system then imagines more views that agree with the skeleton, adding detail and reach. A final rendering method, 3D Gaussian Splatting, then turns the image set into a scene viewable from any angle. Experiments show cleaner textures and more consistent structure than earlier methods, especially far beyond the original view, where rivals tend to fall apart.

Catalog record

CREATORZhexiao Xiong, Zhang Chen, Zhong Li, Nathan Jacobs, [0.5ex] OPPO US Research Center
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-05152

Browse the full library · About the Tower of Babel