abyz22_RemoveControlnet

I'm going to save you the hour I spent: this node does not work. abyz22_RemoveControlnet is supposed to strip ControlNet conditioning out of a conditioning object - the idea being you feed it the base conditioning plus the ControlNet-applied one and get back the clean version. The inputs and output are declared as such. But the shipped code is a debug stub: it prints the conditioning dicts to the console and returns None. ComfyUI expects a value from a node's FUNCTION; returning None fails the run.

I'm telling you this up front because the node renders fine in the graph, looks plausible, and will eat your whole queue at execution. Knowing the pack helps explain it: abyz22/image_control is a single-developer fork of Impact Pack with several half-finished pieces, and this is one of them. The author left the debug prints in - the print statements dump the conditioning dictionaries, which suggests it was mid-development when the pack was published.

What it declares

• Conditioning - the base conditioning.
• ControlNet - the ControlNet-applied conditioning.
• Conditioning (output) - the intended stripped result.

Both inputs are required; the output is declared CONDITIONING. If you wire it up, expect a crash or a broken queue the moment it executes.

The honest alternatives

If you need to remove a ControlNet from conditioning, you don't need this node - re-encode the prompt without applying the ControlNet. In this pack's own toolchain, abyz22_Padding-Image applies a pose ControlNet to conditioning; if you want the un-conditioned version, take the conditioning output from before the ControlNet application or just don't connect the pose branch. ComfyUI's ConditioningZeroOut and the plain CLIPTextEncode give you full control over what ends up in your conditioning without gambling on this stub.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/abyz22/image_control

or ComfyUI Manager → "image_control", restart.

Bottom line

Treat it as documentation of intent, not a tool. If you found a workflow that references it, the workflow is broken unless the author patched it post-crawl. The rest of this pack (the LaMa nodes, wildcard encoder, pipe system) is worth keeping; this particular node you should drag into the trash and re-route around. And if the maintainer ever fixes it, check the console output first - that's where the current version does all of its "work."