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

Temporal Scaling Ablations

We evaluate how generation quality evolves as temporal length increases from 81 to 648 frames at a fixed 2 views (Tab. 5, Sec. 4.2 and Fig. 10 in the paper). Cross-view synchronization remains stable across an 8x increase in duration, while visual quality shows slight degradation for long durations. For each configuration, we show both views. Hover over any video to play.

"A man in a pink t-shirt is practicing martial arts on a wooden deck surrounded by a sandy beach and the ocean with a clear blue sky and palm trees in the background."

81 Frames

162 Frames

324 Frames

648 Frames