Multi-Shot and Scene-Transition Video Generation

Develop text-to-video generation methods capable of synthesizing videos that consist of multiple shots and include explicit scene transitions, extending beyond single-shot clips while preserving visual quality and temporal coherence.

Background

Lumiere focuses on generating single, full-frame-rate clips to achieve globally coherent motion. While effective for single-shot content, the authors note that their approach is not intended for multi-shot videos or for handling scene transitions, identifying this as a limitation.

Addressing multi-shot composition and scene transitions is highlighted as a distinct open challenge for future research, suggesting the need for new generation or composition mechanisms that can stitch shots and transitions cohesively.

References

As for limitations, our method is not designed to generate videos that consist of multiple shots, or that involve transitions between scenes. Generating such content remains an open challenge for future research.

— Lumiere: A Space-Time Diffusion Model for Video Generation  (2401.12945 - Bar-Tal et al., 2024) in Section 6, Conclusion

These cases show that MemComposer improves grounding and sequence structure, but motion realism and precise video-level continuity remain open challenges.

— Memory-Guided B-Roll Generation from User Video Collections  (2610.01884 - Ham et al., 1 Oct 2026) in Section "Qualitative Results," paragraph "Failure modes"

However, controlling the overall pacing/dynamics of the sequence is an open question on its own.

— Memory-Guided B-Roll Generation from User Video Collections  (2610.01884 - Ham et al., 1 Oct 2026) in Appendix, Section "Limitations"