OpenPoseDepth

OpenPoseDepth is the ComfyUI-CUP node that loads depth control images into your graph - the grayscale maps where brightness encodes distance, so the model knows what's in front and what's behind. In a 3D animation pipeline that's a gift: Blender has the actual scene geometry, so its depth maps are exact in a way that a monocular depth estimator like MiDaS or Depth Anything can only approximate. You get pixel-accurate spatial control for free, straight from the renderer.

Yes, the "OpenPose" in the name is misleading. This whole pack is named around pose, but the nodes are a family of frame loaders, one per control type, and this one's control type is depth. Like its siblings, OpenPoseDepth is an OpenPoseBase subclass with posetype = "depth" - no detection, no estimation, just loading <image-dir>/depth/ImageNNNN.png.

How it works

• image (STRING) - directory of rendered control images, filled by the Blender addon.
• frame (INT, default 0) - picks the file: frame 0 → Image0000.png, etc.

Outputs are IMAGE (the depth map → wire into Apply ControlNet with a depth ControlNet model) and MASK (from alpha, optional). The fun part of using real rendered depth: because the map is geometrically true, you can do things like rack focus or composite layers with confidence, and depth-conditioned generation stays glued to the scene's actual structure instead of the model's guess about it.

Install and gotchas

Part of ComfyUI-CUP - Manager search ComfyUI-CUP, or:

cd ComfyUI/custom_nodes
git clone https://github.com/AIGODLIKE/ComfyUI-CUP.git

Restart after installing. No models to download, no extra dependencies.

The standard CUP warnings apply: this node doesn't compute depth (Blender does), and it's inert without the ComfyUI-BlenderAI-node addon driving it. And a missing frame returns a silent 64×64 black placeholder rather than an error - a black depth map fed to ControlNet will flatten or wreck the output. Check that depth/ImageNNNN.png exists before you go hunting for sampler bugs.