: LCT expands the context window of video diffusion models, allowing them to maintain visual and dynamic consistency across multiple different shots within a single scene.
[2503.10589] Long Context Tuning for Video Generation - arXiv lu-VEnnv2a4.mp4
: The model can extend a single shot to minute-long durations by auto-regressively generating 10-second segments that seamlessly connect without visible cuts. : LCT expands the context window of video
: It facilitates a "director-like" workflow where users can progressively develop content shot-by-shot, using previously generated footage as a reference for the next segment. lu-VEnnv2a4.mp4
According to the research published on arXiv , the key "long features" enabled by this model include: