Search papers, labs, and topics across Lattice.
This paper introduces DramaDirector, a geometry-guided framework for generating short dramas that effectively integrates global plot structures with local context to create visually grounded multi-shot videos. By decoupling static visual elements from dynamic narrative conditions and employing a learned text-visual alignment reward, DramaDirector enhances the generation process through schema-constrained supervised fine-tuning and guided retrieval of cinematographic geometry. Experimental results demonstrate that DramaDirector significantly outperforms existing multi-agent and video generation baselines in terms of faithfulness, consistency, and controllability, while also establishing a comprehensive benchmark with DramaBoard for evaluating drama generation.
DramaDirector achieves unprecedented fidelity in short drama generation by leveraging real cinematographic geometry, outperforming traditional video generation methods.
Short dramas, with their rapid shot rhythms, dialogue-driven focus shifts, and demanding cinematographic grounding, pose challenges that prompt-level or text-only video generation pipelines struggle to meet. We study plot-to-short-drama generation, where a global plot and local context are transformed into visually grounded multi-shot videos. We propose DramaDirector, a geometry-grounded framework that lets the planner borrow cinematographic geometry from a gallery of real short-drama shots indexed by depth and pose. DramaDirector decouples each shot into static visual and dynamic narrative conditions, trains the planner with schema-constrained SFT and GRPO under a learned text-visual alignment reward, and retrieves depth-pose references to guide first-frame generation and image-to-video synthesis. We also introduce DramaBoard, a benchmark built from 35 live-action dramas, 2.8K episodes, and 81K shots, with structured storyboards and multi-dimensional evaluation protocols. Experiments show that DramaDirector improves over representative multi-agent and video generation baselines on faithfulness, consistency, and controllability. Our code is released at: https://github.com/iLearn-Lab/DramaDirector
