Submitted by KAI LIU 19 JavisGPT: A Unified Multi-modal LLM for Sounding-Video Comprehension and Generation JavisVerse 68 3