Run SAM 1/2/3, EfficientSAM, YOLO-World and other promptable vision models locally, on CPU, CUDA or CoreML.