OpenOutpainter Serving Output Upscale

The last stop on the upscale route. Your graph took the image from the Input Upscale node, ran it through your upscaler of choice, and now this output node sends the finished result back to openOutpaint so the canvas can replace its version with the higher-resolution one. It's the smallest of the pack's output nodes, and the one with the fewest surprises.

Two required inputs, no outputs - it's a terminal node:

• oop_request - the request object, threaded through from the Serving node.
• image (IMAGE) - the upscaled result from your upscaling branch.

How it works

Command-aware and lazy, same as every node in this family. It only acts when the pending request is an upscale call (/sdapi/v1/extra-single-image/) and image is present. When it fires, it base64-encodes the image and finalizes the request with the response body:

{"image": "<base64>"}

The HTTP handler that's been blocked since the canvas POSTed wakes up, writes that JSON back, and the canvas swaps in your upscaled image. Note there's no SEEDS input here, unlike the txt2img/img2img output nodes - upscaling doesn't carry seeds, so the response is just the one image.

The one thing to get right is the image wire. It's lazy, so if it's not connected to a real IMAGE output, the node simply never runs and the canvas hangs with no error. Same story if the request is a different command type - this output only fires for upscale, and pairing it with the wrong input node means the branch never activates.

Install and the practical stuff

The pack installs once for all of its nodes: ComfyUI Manager (search "OpenOutpaint ComfyUI Interface") or git clone https://github.com/without-ordinary/openoutpaint_comfyui_interface into ComfyUI/custom_nodes, then restart. Dependency-wise it's just opencv-python; no model files to fetch.

Practical gotchas are the pack's standard ones: an error upstream in your upscaler leaves the canvas request hanging until you re-run the workflow with the same request_id, and a second Serving node in the graph is unsupported and can steal requests. Also keep in mind the canvas's upscaler dropdown is placeholder-driven ("[WIP] Use Workflow"), so whatever the user selects there, the actual upscale quality is whatever your graph does with the image this node returns.