CtrlNet MiDaS Pipe (JPS)

The other half of CtrlNet MiDaS Settings (JPS). That node bundles six MiDaS depth-ControlNet values into one wire; this one unpacks the bundle wherever downstream you actually need them. Same trick as every Settings/Pipe pair in this pack - one menu, placed once, and a thin pipe node dropped wherever the graph needs the individual values.

What comes out of it

One input, midas_settings (a BASIC_PIPE - JPS's own bundle, a plain tuple under the hood, not Impact Pack's object despite the shared type label). Six outputs:

• midas_source (INT) - which reference image to run MiDaS on. It leaves as a plain integer, not the "Source Image / Support Image / Support Direct" text - that matches the select input on the pack's Image Switch node, so the obvious wiring is straight into one.
• midas_strength (FLOAT) - feeds your ControlNet Apply node's strength widget.
• midas_start / midas_end (FLOAT) - the step-window percentages for Apply ControlNet (Advanced)'s start/end inputs.
• midas_a (FLOAT) - MiDaS's own tuning parameter for its depth-to-normal math, feeds the MiDaS preprocessor node's matching input.
• midas_bg (FLOAT) - the background threshold, also feeds the MiDaS preprocessor's matching input.

This node doesn't estimate depth or apply anything - it's routing, full stop. The actual work is a MiDaS depth preprocessor node upstream (turning your source image into a grayscale depth map) and an Apply ControlNet node downstream (conditioning generation on it); this pipe just carries the right number to each.

Installing it

Comes bundled with the pack. ComfyUI Manager: search JPS Custom Nodes for ComfyUI, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/JPS-GER/ComfyUI_JPS-Nodes.git

Restart afterward, and delete any prior install of the pack first - the README is explicit that leftover old copies cause problems. Nothing to download here; the MiDaS model weights themselves live with whatever preprocessor node actually runs the depth estimation, not with this routing node.

Where people get tripped up

Without a CtrlNet MiDaS Settings (JPS) node feeding it, this node has nothing to unpack - it's not usable standalone. And it's easy to wire the strength/start/end outputs correctly and forget midas_a/midas_bg entirely, since those two only matter if your MiDaS preprocessor node happens to expose matching inputs - not every depth preprocessor does, so check what you're actually running upstream before assuming these two outputs are doing anything. Like its siblings across this pack's ControlNet family, it's newer than the published README and won't show up in that document's node list.

Worth a mental note if you're building a workflow that needs to swap between different depth estimators later: this node's outputs are named specifically for MiDaS (midas_a, midas_bg don't mean anything to a Depth Anything or Marigold preprocessor). If you upgrade the actual depth model down the line, you'll be replacing this Settings/Pipe pair too, not just the preprocessor node - the settings schema is tied to which estimator it was built for.