Text-Driven Artistic Staging: 3D Posing, Lighting, and Camera References from Paintings
This episode explores Yunge Wen's paper on text-driven artistic staging. The model uses a dataset of 2,328 paintings to unify 3D posing, lighting, and camera placement from text prompts. By learning from historical art, it achieves higher retrieval accuracy than CLIP and offers diverse interpretations through flow-matching transformers.