LLaVA-OneVision 1.5 Fully open multimodal (images/video+text) models & training stack; strong results and reproducible recipes. 0430 Open-source Models# LLaVA# local# multimodal