Multi-Grid Post-Training for Long-Form Multi-Shot Video Generation
MovieGrid turns long videos into a grid of shorter ordered chunks so the model can keep more shots coherent under the same token budget.
The paper says this avoids forcing an entire narrative down one temporal axis, while still letting chunks exchange global context.
Its MGLV dataset contains 54K grid videos built from 1,000 long-form videos with character-aware story prompts.
The authors report 6.05x more shots than Temporal Packing in a 1,616-frame video, plus stronger intra-shot and inter-shot consistency than named baselines.
Source: HF Daily Papers' note.
The paper says this avoids forcing an entire narrative down one temporal axis, while still letting chunks exchange global context.
Its MGLV dataset contains 54K grid videos built from 1,000 long-form videos with character-aware story prompts.
The authors report 6.05x more shots than Temporal Packing in a 1,616-frame video, plus stronger intra-shot and inter-shot consistency than named baselines.
Source: HF Daily Papers' note.
score 5