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

Supplementary Material

Please click on the sections below to view our interactive results, comparisons, and ablations:

• Our Results
• Real-World Results
• Short Sequence Comparisons
• Long Sequence Comparisons
• Component Ablations
• View Scaling Ablations
• Temporal Scaling Ablations

Long Sequence Comparisons

We compare MV-Forcing against SF+ReCamMaster and SF+ReCamMaster+SF on long multi-view video generation (Tab. 2, Fig. 3, Sec. 4.1 in the paper). For each scene, we show all 3 views generated by each method. All methods operate on real-world data. SF+ReCamMaster generates each view as independent 81-frame segments with no shared temporal context, leading to visible degradation beyond the first segment. SF+ReCamMaster+SF recovers temporal continuity but lacks cross-view conditioning, resulting in geometric drift between viewpoints. Our method maintains consistency across both views and time via the accumulated CUT3R geometric prior and spatio-temporal self-forcing. Hover over any video to play.

"A young man wearing white clothes practicing karate kicks on an urban street."

SF+ReCamMaster

SF+ReCamMaster+SF

Ours