LanguageBind / Video-LLaVA

The foundational multimodal line of Li Yuan's PKU-YuanGroup: LanguageBind (ICLR 2024) binds video, infrared, depth, and audio to a frozen language encoder; Video-LLaVA (EMNLP 2024) unifies image and video instruction tuning on the aligned representation and became a standard open video-LLM baseline. The lineage (with MoE-LLaVA and MagicTime) seeded the group's LanguageBind HF org and led to LLaVA-CoT (ICCV 2025, 2.1K★), the first widely-noted open "visual o1" structured-reasoning VLM.