CMK Pipe Set Detailer Result

The detailer returned an IMAGE, and the pipe has no idea about it yet. This node fixes that: pipe plus image in, pipe out, with the image stored under image_detailer.

Two keys, actually, and the second is the interesting one. It also writes image_final.

That second write is documented in the source as a staging fallback for workflows without a later face module. Which is a tidy way of saying: if your graph is detailer and then straight to saving, the detailer's output is the last image you're going to get, so it should be the one the output stage picks up. Wire in a FaceProcess or FaceRebuild module after this and that later module writes its own image_final, correctly overriding the staging value. The pack is arranging for the right image to be final without you having to think about which module happens to be last today.

Notice what it does not touch: image. The original generation image stays in the pipe as image, so anything downstream that wants the pre-detail version - a comparison, a diagnostic, a second branch - still has it. Overwriting image here would quietly destroy the reference the whole pack uses for its before/after visualisations.

When you'd use it

Mainly two situations. Building or modifying a CMK module by hand, where you're wiring the detailer stage's output back onto the bus yourself. Or extending a flow: you did something to the image outside the pack - an upscale, an inpaint, a colour fix - and you want the downstream CMK stages to treat that as the detailed result.

If you're using the packaged SMART DETAILER/DETAILER SDXL modules unmodified, this node is already inside them and you never see it. Same as every CMK/Developer/ node in this pack: it's the module's internal plumbing, exposed in the node menu because ComfyUI subgraphs can't hide it, not because you're meant to reach for it.

Installing the pack

ComfyUI Manager → search CMK Flow → install the entry published by CMKFlow → restart ComfyUI. The registry name is cmk-flow.

Manual install from the ComfyUI root directory:

git clone https://github.com/CMKFlow/cmk_nodes.git custom_nodes/cmk_nodes
python3 custom_nodes/cmk_nodes/scripts/install_cmk_requirements.py

You're looking for CMK dependency check: OK on the last line. Do not skip that second command - a git clone never runs pip, and the pack's requirements.txt (ultralytics, onnxruntime, insightface, segment-anything, opencv-python, imageio-ffmpeg, nudenet==3.4.2) is what makes the detailer, detection and face modules importable at all. The insightface build is the one that fails most often on a bare machine; the installer at least tells you clearly instead of half-loading the pack.

Because the pack imports all of its nodes in one graph, a dependency that fails to build doesn't break one node - nodes go missing from the menu. If CMK nodes are absent entirely after an install, read the ComfyUI console startup log before touching the workflow.

Two interface requirements: Vue Nodes / Nodes 2.0 must be on, and Execution → Live preview method should be auto if you want previews while the sampler and refiner run.

Also: don't copy the JSON blueprints from subgraphs/ into user/default/subgraphs/. ComfyUI registers node-pack blueprints and user blueprints as separate sources, and duplicating them gives you double entries in the browser.

Small print worth reading once

The stale-context rule applies here as it does to every bundle node. This write is a snapshot; change what feeds image and the pipe keeps the old value until this node re-executes. First run after a restart or cache clear logs MISS → STORED, then HIT for unchanged repeats within the session - CMK's image boundaries live under ComfyUI/temp/cmk/ and are cleared on restart. That's expected behaviour, not a broken cache.