This research project introduces a framework for segmenting any object in 3D scenes represented as neural radiance fields (NeRFs). Users provide a single manual prompt from one rendered view, and SA3D propagates the segmentation across the entire 3D model. The approach leverages the Segment Anything Model (SAM) for 3D scene perception and achieves results in approximately 2 minutes without engineering optimization. The work was published at NeurIPS 2023 and IJCV 2025.