Zhaoqi Wang, Zijian Zhang, Jiamou Liu
SpriteStage enables long-form animation generation by iteratively executing and refining scenes within a shared virtual stage, using runtime feedback to adjust narrative and action elements. It combines scene programming with engine reorchestration to align generated content with production-ready assets while preserving timing and event continuity. The framework produces coherent, extended animations up to 14.3 minutes in length.
Long animations require consistent narrative and visual coherence across scenes, which current video generation models struggle to maintain. SpriteStage addresses this by enabling dynamic, feedback-driven refinement of scenes during execution, making it feasible to generate complex, production-quality animations without predefining the entire story upfront.
Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.