Depth View to Isometric

A raw depth map is one of the least satisfying things to look at in ComfyUI. You run Depth Anything V2, you get a grayscale image where white is close and black is far, and you're supposed to take its word that it's 3D. Visually it isn't. This node is the part that finally makes it look 3D.

DepthViewToIsometric takes the pack's custom DEPTH_VIEW type - the geometry object that DisparityToDepthView produces - and renders it as an isometric projection. The depth becomes a heightfield, the camera tilts up to your chosen elevation and swings around to your azimuth, and you get an orthographic, no-perspective view of the extruded surface, colored by your albedo image and optionally shaded by a normal map. Seen those "living photo" parallax demos? This is the same geometry, frozen and inspectable instead of wiggling.

Why you'd reach for it: to actually check your depth before you commit to a parallax or VR pipeline. A depth map can look fine as a gray blob and be a mess as geometry (or vice versa). This render shows you folds, holes, and inverted regions at a glance. It's also the fastest way to fake a 3D thumbnail of a 2D scene - and it doubles as a bas-relief preview, since an orthographic render of a heightfield is basically what a 3D printer sees.

The inputs that matter:

• depth_view - the DEPTH_VIEW type from DisparityToDepthView in this same pack. A regular IMAGE depth map won't plug in; the type is specific to this pack.
• albedo - the color image that paints the surface. Your original photo, or whatever you generated.
• elevation / azimuth - where the virtual camera sits. Elevation goes -90..90 (default 15), azimuth -180..180 (default 30). These two are the fun part: spin the view until the geometry reads clearly.
• camera and axis - booleans that toggle helper overlays in the render, a camera marker and the XYZ axes. Axis defaults on, so flip it off before you save a clean output.

The optional normals input comes from DisparityToDepthView and gives the surface real shading - without them you get flat albedo dumped on a heap; with them, the folds and contours read properly. The single output is an IMAGE you can preview, save, or pipe anywhere.

Install it via ComfyUI Manager → Install Custom Nodes, searching comfyui-sunflower-nodes, then restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/andygill/comfyui-sunflower-nodes

Then restart ComfyUI. There are no model downloads and no extra pip packages - every node in this pack is pure numpy/torch geometry math, nothing beyond what ComfyUI already ships. One honest wrinkle: the repo has gone quiet since spring 2025 and the GitHub page is currently unreachable, so if the clone 404s that's the repo, not you - a cached or archived copy installs the same way.

Troubleshooting is mostly wiring. A black or empty output almost always means depth_view isn't connected (or is stale from an earlier run). If the geometry renders upside-down, your disparity polarity is inverted upstream - flip the mask before it gets here. And if you keep seeing axes in your "final" image, you forgot the axis toggle. The name is a lie only insofar as it implies an API or a model - there's neither. It's just math, and it's the reason this pack exists.