Walk→3D turns a short walkthrough video into a colored 3D point cloud. It samples sharp frames across the video, uses the open-weight Depth Anything 3 (DA3-BASE) model to predict depth, confidence, and camera parameters, then combines confident pixels into a shared 3D scene. The project exports a PLY point cloud and camera path, and includes a Gradio interface with an interactive 3D viewer. This is an early prototype: results can contain missing geometry or camera drift, and have not yet been validated against a ground-truth scan.