Comfy plugin

[clear plaza]

anyway to use direct ip within lan with comfy plugin? I have openclaw running in a weak box and want it to call the comfy instance that runs on my gaming pc. what are possible values of comfy.config.mode?

[white laurel]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[remote halo]

Yep — use mode: "local" with a LAN baseUrl pointing at the gaming PC.

Possible plugins.entries.comfy.config.mode values are only:

• "local" — direct ComfyUI API, defaults to http://127.0.0.1:8188
• "cloud" — Comfy Cloud, defaults to https://cloud.comfy.org

Example:

{
  plugins: {
    entries: {
      comfy: {
        config: {
          mode: "local",
          baseUrl: "http://192.168.1.50:8188",
          image: {

            workflowPath: "./workflows/flux-api.json",
            promptNodeId: "6",
            outputNodeId: "9",
          },
        },
      },
    },
  },
}

On the gaming PC, make sure ComfyUI is listening on the LAN interface, not only localhost — usually start it with something like --listen 0.0.0.0, and allow port 8188 through the firewall.

Docs: https://docs.openclaw.ai/providers/comfy

[clear plaza]

so with IP, the workflowPath should relative to the gaming pc? or I must use local json file

[remote halo]

workflowPath is read by OpenClaw on the machine running the gateway/plugin — your weak box, not the gaming PC.

The IP/baseUrl is only where OpenClaw submits the workflow to ComfyUI. The workflow JSON itself must be available to OpenClaw, either as:

image: {
  workflowPath: "/home/you/openclaw-workflows/flux-api.json",
  promptNodeId: "6",
  outputNodeId: "9",
}

or inline as workflow: { ... } in config.

So: export the API workflow JSON from ComfyUI, copy it to the OpenClaw box, and point workflowPath there. Comfy on the gaming PC still needs the actual model/custom nodes/files installed locally, since it’s the one executing the graph.

[clear plaza]

with comfy can we use image editing? how can we pass image as input to workflow?

[remote halo]

Yes — Comfy image editing is supported, with one uploaded reference image.

How it works:

• You configure an inputImageNodeId in the image workflow config.
• When the user/tool call includes an input image, OpenClaw uploads it to Comfy via /upload/image.
• Then OpenClaw writes the uploaded filename into that node’s input, default input name image.

Example:

plugins: {
  entries: {
    comfy: {
      config: {
        mode: "local",
        baseUrl: "http://192.168.1.50:8188",
        image: {

          workflowPath: "./workflows/edit-api.json",
          promptNodeId: "6",
          inputImageNodeId: "7",
          inputImageInputName: "image", // default is "image"
          outputNodeId: "9",
        },
      },
    },
  },
}

So your workflow needs a node that normally receives an uploaded image filename — often a LoadImage-style node. Set inputImageNodeId to that node ID, and inputImageInputName to the exact input field name in the API JSON.

Current limitation: only one reference image is supported for image workflows. Docs: https://docs.openclaw.ai/providers/comfy

[clear plaza]

can I have 2 workflows, 1 for txt2img, 1 for img editing?