OpenOutpainter Serving Output TXT2IMG

The bookend to the txt2img input node. Your graph has generated the images and the openOutpaint canvas is sitting on the other end of an HTTP request, waiting. This output node is what closes the loop: it takes the finished IMAGE tensors, base64-encodes them, wraps them in the exact JSON shape the A1111 API would have returned, and unlocks the waiting request. The canvas stamps the result in place and you keep painting.

It's an output node, so it has no output sockets of its own - it's the end of the branch. It takes three required inputs:

• oop_request - the request object, passed down from the Serving node through the txt2img input node.
• images (IMAGE) - whatever comes out of your final VAE Decode.
• SEEDS (INT) - the seed(s) you actually used. It's a forced input, meaning you wire it rather than typing it, and it's treated as a list - if batch_size was 3, you should be feeding three seeds.

How it works

Like the input nodes, this one is lazy and command-aware. It only does anything when the pending request is a /sdapi/v1/txt2img call, and only when images and SEEDS are both present. When it fires, it builds the response:

{"images": ["<base64>", "..."], "info": {"all_seeds": [123, 124, 125]}}

info.all_seeds mirrors A1111's response so openOutpaint can display which seeds produced which image. Then it calls finalize() on the request, which sets a threading event - the HTTP handler in the Serving node has been blocked on that event since the canvas POSTed, and now it wakes up, writes the JSON response, and cleans the request out of its queue.

The images wire is the one to get right. It's lazy, so a wrong connection usually shows up as "the canvas hangs forever" rather than an error. If images never arrives, the node stays dormant and the POST never completes.

Install and the hang problem

Install comes from the pack as a whole - ComfyUI Manager (search "OpenOutpaint ComfyUI Interface") or git clone https://github.com/without-ordinary/openoutpaint_comfyui_interface into ComfyUI/custom_nodes, restart, done. No models, and opencv-python is the only pip dependency worth noting.

The failure mode to know before you hit it: if anything in the graph before this node errors out, the canvas request hangs with no error shown anywhere - the code comment says to just run the workflow again with the same request_id to complete the API call. That's not a great experience, but it's the honest price of a graph running as an API server. And remember the pack-wide constraint: only one serving workflow at a time, so if this node's branch never seems to get the right oop_request, check you don't have a second Serving node competing for requests.