Extract Depth Channel

A cubemap with depth is two things stuck together: the RGB faces and a separate depth layer per face. Extract Depth Channel pulls that depth layer out and hands it back as its own cubemap with depth rendered as grayscale RGB - so you can look at it, save it, or flatten it to a depth panorama.

Why would you bother? Because depth is invisible inside the CUBEMAP type. If you've run Batch Cubemap Depth and converted straight back to equirect, you've seen the RGB result - but you've never actually verified what the depth map looks like until you render it. This is the node for that verification pass: wire it in, pipe the output to Cubemap Preview with show_depth on, and suddenly the "depth" your model produced is right there in grayscale where you can check whether the sky is cleanly black and the foreground pops.

Mechanically it's a channel copy. Each face's single-channel depth gets stacked into three identical channels, producing a grayscale visualization. The output is still a CUBEMAP - just one whose RGB content is the depth picture - which means the whole rest of the pack can keep working on it. Flatten it with Cubemap to Equirect using the depth output type and you've got a depth map of the full panorama, not just the faces.

Inputs and outputs

• cubemap - any cubemap that has depth (anything that's been through Batch Cubemap Depth, Apply Depth to Cubemap Face, or Merge Cubemap Depth).

Output is a CUBEMAP where the faces are the depth rendered as grayscale RGB. It will throw if the cubemap has no depth yet - that's not a bug, that's the pack telling you you hooked it up before any depth existed.

How to install it

ComfyUI Manager, search "DreamCube", install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/cedarconnor/ComfyUI-DreamCube.git
cd ComfyUI-DreamCube
pip install -r requirements.txt

No models, no keys. All the heavy lifting is NumPy/SciPy, which ComfyUI already ships.

Gotchas

Don't confuse this with Merge Cubemap Depth, which does the opposite direction (RGB + depth → RGBD). This node is strictly for extracting a viewable depth cubemap. And because it's a visualization pass, the depth is now baked into RGB channels - don't try to run more depth processing on the output and expect it to act like a depth layer again. Extract, look, flatten, done.