The researchers introduce ContextMaster, a unified framework designed for Interactive Multi-Shot Video Creation (IMVC), a setting that requires seamless transitions between text-to-video generation, reference conditioning, and video editing while maintaining historical consistency. To prevent the computational overhead of an expanding context history during the denoising process, the authors implement a fixed-budget sparse context routing mechanism. This method utilizes reusable clean context states and a specialized "ConstraintSink" component to ensure task constraints remain visible despite the sparse access. To optimize the model for efficient inference and few-step denoising, the authors developed a two-stage privileged context distillation framework. This process first employs consistency distillation to transfer dense teacher context behavior to the sparse model, followed by distribution matching to refine deployment rollouts. Experimental results across three primitive video tasks demonstrate that ContextMaster achieves superior task fulfillment and inter-shot consistency compared to specialized baselines. Furthermore, the model maintains high computational efficiency, reaching an inference speed of 16 FPS on a single GPU, while user studies confirm the effectiveness of its flexible, composable workflows for complex video production tasks.