Proact-VL：用于实时 AI 伴侣的主动视频语言模型

Proactive and real-time interactive experiences are crucial for human-like AI companions, yet they encounter three main challenges: achieving low-latency inference under continuous streaming inputs, autonomously deciding when to respond, and controlling both the quality and quantity of generated content to meet real-time constraints. This research instantiates AI companions through two gaming scenarios, commentator and guide, selected for their suitability for automatic evaluation. A Live Gaming Benchmark is introduced, comprising a large-scale dataset with three representative scenarios: solo commentary, co-commentary, and user guidance. Proact-VL is presented as a general framework that shapes multimodal language models into proactive, real-time interactions.