MV-Forcing: Long Multi-View Video Generation via 4D-Grounded Spatio-Temporal Self-Forcing

2026-07-06

AI Summary

Existing methods struggle to generate temporally long and multi-view geometrically consistent videos of dynamic scenes. This work proposes a 4D geometry-guided spatiotemporal autoregressive diffusion model that integrates temporal and view autoregressive mechanisms, coupled with joint denoising training and distribution-matching distillation to effectively mitigate exposure bias between training and inference. Leveraging 3D reconstruction-derived geometric priors, the model enables unbounded temporal length and arbitrary numbers of views within a single unified framework while preserving geometric consistency. Experiments demonstrate that high-quality, geometrically coherent multi-view dynamic videos can be synthesized on both synthetic and real-world datasets using only a few sampling steps.

Abstract

Recent advances in video diffusion models have enabled either long single-view generation through temporal autoregression, or short multi-view synthesis through bidirectional attention. However, generating long, multi-view consistent videos of dynamic scenes remains unsolved. In this work, we present MV-Forcing, a framework that composes temporal and view-wise autoregression within a single diffusion model by introducing a 4D geometric bridge between sequentially generated views. Our key insight is that an autoregressive 3D reconstruction model naturally interfaces between autoregressively generated views. Given a completed source view, we reconstruct its 3D structure and render a geometric prior of the next target viewpoint, which the diffusion model refines into a high-quality video. To extend generation beyond the teacher's fixed temporal window, we introduce a joint denoising regime where both view slots are initialized from noise during training, enabling temporally unbounded generation. We distill the model via Distribution Matching Distillation with Spatio-Temporal Self-Forcing, closing the train-inference exposure bias gap for both temporal and view-sequential autoregression. Extensive experiments on both synthetic and real-world data demonstrate that MV-Forcing produces geometrically consistent multi-view videos of dynamic scenes at arbitrary lengths and viewpoint counts using a single few-step student model.

Problem

• multi-view video generation
• long video synthesis
• temporal consistency
• geometric consistency
• dynamic scenes

Innovation

• MV-Forcing
• 4D-Grounded
• Spatio-Temporal Self-Forcing
• Multi-View Video Generation
• Autoregressive Diffusion