Which base model does it really descend from?
Query a phone-video 3D scene in natural language — no GPU.