Bridge Input (IMAGE)

Bridge Input (IMAGE) is where the scalar bridges stop and the complex-type ones start. STRING, INT, FLOAT, BOOLEAN all have a widget you can type into. IMAGE has no widget at all - an image isn't a number you patch into a JSON field, it's a tensor that lives inside ComfyUI. So this node's job is different: it's a named tap on an image stream, a marked place in the workflow where an agent knows "the source image flows through here."

Here's the honest picture of how ComfyBridgeIO handles this. The node has an optional value input socket (IMAGE) and a single value output socket (IMAGE). You wire a Load Image (or any image-producing node) into the input, then take the output to wherever the graph needs the image. When the Python client runs the workflow it leaves the image pipeline alone - it patches the scalar Bridge Inputs and the name tags, and the image just flows. The classic setup: the workflow already has a Load Image wired into this bridge, and the agent changes which image by uploading a file (bridge.upload_image(...) returns a server filename) and patching the loader. The bridge itself is the stable landmark.

The name widget (STRING, default "input_name") is still the contract - it's how the client reports and matches this input, and what extract_schema() surfaces so the agent knows an image input exists at this spot.

Install

Same as every node in this pack:

cd ComfyUI/custom_nodes
git clone https://github.com/cbpark84/ComfyBridgeIO

Restart ComfyUI, or use ComfyUI Manager → search ComfyBridgeIO → Install. No models, no heavy deps. Optional Python client via pip install -e in the repo (aiohttp, websockets, Pillow). The README's clone line says nicekate/ComfyBridgeIO - stale; the repo is cbpark84 as above.

Common issues

The trap is expecting the client to conjure an image tensor out of thin air. It can't - IMAGE values in API-format JSON are node links, not raw pixels. If nothing is wired into this bridge's input, its output is empty and everything downstream silently gets nothing. Build the image path inside ComfyUI (Load Image → this node → the rest), and let the bridge be the named hole the agent targets. That's the pattern this pack actually supports.