ComfyUI Client Get Output (Yogurt Nodes)

ComfyUI Client Run hands you a bag of everything the remote workflow produced. That bag is useless until you open it - which is this node's entire job. ComfyUI Client Get Output takes a COMFYUI_RESULTS package plus a node id or title, and returns that one node's outputs as data. It's the final step in the eight-node ComfyUI Client family: Load configures, Set nodes patch, Run submits and waits, Get Output digs out what you actually wanted.

How it works

When a ComfyUI server runs a workflow, it returns a results structure keyed by node - each node's outputs live under its id. Get Output looks that up for you. You give it the node field as either the numeric id (the number you see in the workflow JSON) or the node's title (the human name you gave it in the remote graph), and it returns everything that node emitted as a single data value of any type. So if the remote workflow ends in a Save Image node and you want the image tensor itself, you ask for that node by title and get the tensor back, ready to wire into your local graph.

There's a convenience shortcut baked in: leave node blank, and if the results package contains exactly one node's output, it auto-selects it. Handy in simple one-output workflows, and a trap in big ones - the moment the remote produces output from several nodes, a blank field has nothing to auto-pick and you get nothing useful.

Inputs

• results - the COMFYUI_RESULTS package from Run.
• node - the node id or title to extract. Blank = auto-pick if there's exactly one output node in the package.

Outputs

• data - the selected node's full output list, as an any-typed value. This is what flows back into your local graph: an image, a string, a mask, whatever the remote node emitted.

Install

Standard Yogurt Nodes install. ComfyUI Manager → search "ComfyUI-YogurtNodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/yogurt7771/ComfyUI-YogurtNodes.git
cd ComfyUI-YogurtNodes && pip install -r requirements.txt

Then restart ComfyUI. No dependencies beyond the pack base - this is pure data extraction, no network calls of its own.

Where people get burned

The #1 gotcha is stale references. The node you name must match the remote workflow's ids/titles at run time, and the remote workflow can change under you. You set node to "3" because that was the Save Image node last week; the remote graph got rearranged and now "3" is something else entirely - you'll get the wrong data, silently, with no error. Titles survive renumbering better than ids, so prefer titles when the remote is being actively edited, and re-export the workflow whenever it changes. Second, remember the blank-field auto-pick only fires for a single output node. If the remote produces several outputs and you leave it blank, you'll stare at an empty result and wonder why - the answer is always "be explicit about the node." And third: data is a list of the node's outputs, not a single value. If the remote node outputs one image you get a one-item list; downstream nodes expecting a bare image tensor may need the unpack. When the wire won't connect, that's usually it.