Stepper: Stepwise Immersive Scene Generation with Multiview Panoramas

View PDF HTML (experimental)

Abstract:The synthesis of immersive 3D scenes from text is rapidly maturing, driven by novel video generative models and feed-forward 3D reconstruction, with vast potential in AR/VR and world modeling. While panoramic images have proven effective for scene initialization, existing approaches suffer from a trade-off between visual fidelity and explorability: autoregressive expansion suffers from context drift, while panoramic video generation is limited to low resolution. We present Stepper, a unified framework for text-driven immersive 3D scene synthesis that circumvents these limitations via stepwise panoramic scene expansion. Stepper leverages a novel multi-view 360° diffusion model that enables consistent, high-resolution expansion, coupled with a geometry reconstruction pipeline that enforces geometric coherence. Trained on a new large-scale, multi-view panorama dataset, Stepper achieves state-of-the-art fidelity and structural consistency, outperforming prior approaches, thereby setting a new standard for immersive scene generation.

Comments: Accepted at CVPR 2026 Findings; Find our project page under this https URL

Subjects:

Computer Vision and Pattern Recognition (cs.CV)

Cite as: arXiv:2603.28980 [cs.CV]

(or arXiv:2603.28980v1 [cs.CV] for this version)

https://doi.org/10.48550/arXiv.2603.28980

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Felix Wimbauer [view email] [v1] Mon, 30 Mar 2026 20:26:28 UTC (19,710 KB)