Refresh Remote Models

This one's the sidekick, not the star. It's paired with Remote Upscale, the node in this same pack that offloads the actual upscaling work to a second machine over HTTP. That node has a model_name dropdown it fills in by asking a remote server what's available - and if you add a new upscale model to that server after the fact, or you're just not sure the connection is even working, Refresh Remote Models is what you run to find out.

It's about as small as ComfyUI nodes get: one input, one output, no image processing, nothing generative happening at all. It exists purely so you're not guessing whether your two-machine setup is actually wired up correctly.

How it works

Per the README, this node hits the server's /models endpoint - the same one Remote Upscale's dropdown draws from - and reports what came back. Point it at your server, run it, and you know two things immediately: whether the server answered at all, and what upscale models it currently has loaded from its UPSCALE_MODELS_PATH folder. If you've been staring at a Remote Upscale node stuck on the (server unavailable) placeholder, this is the node that tells you whether the problem is the connection or the model folder.

The inputs and outputs that matter

There's genuinely just one of each:

• server_url - the only input. Defaults to http://localhost:8189, same as its sibling node, and same caveat: if your upscale server is running on a different machine (which is the entire reason this pack exists), you need to point this at that machine's address, not leave it on the default.
• Output - a single STRING. There's no fancy typed response here, just text back from the node; feed it into a Show Text / preview-style node if you want to actually read it on the graph instead of just checking that the run didn't error out.

How to install it

It ships in the same pack as Remote Upscale, so there's no separate install step - get the pack once and you have both nodes.

cd ComfyUI/custom_nodes
git clone https://github.com/gen1nya/ComfyUI-Remote-Upscale

Restart ComfyUI, and you'll find it under image/upscaling alongside Remote Upscale. The client-side pack only needs the requests library - nothing heavy to install here. What does need setting up is the server it talks to: that's a separate FastAPI app (the server/ folder in the repo) that runs on your other machine, needs its own pip install -r requirements.txt, a matching PyTorch build for your GPU, and an UPSCALE_MODELS_PATH pointing at a folder of .pth/.pt/.safetensors upscale checkpoints. See the Remote Upscale article for the full server setup - this node is only useful once that server exists.

Common issues & troubleshooting

It runs but you get nothing useful back. If the server isn't started yet, or server_url is wrong, you'll get an empty or error-ish STRING rather than a model list - that's your signal to go check the server side first, not this node.

Localhost by default. Same trap as the main node: the default server_url assumes the server is on the same machine as ComfyUI. If you're running this against a genuinely remote box, you have to change it - nothing here auto-discovers the server on your network.

You added a new model file and it's not showing up. Make sure the file actually landed inside the folder you set as UPSCALE_MODELS_PATH on the server, and that it's one of the three supported formats (.pth, .pt, .safetensors). This node reports what the server currently sees - it can't reach into a folder the server was never told about.

Wondering why this needs to be a separate node from Remote Upscale at all. It's a fair question for a node this small, but it makes sense once you think about the two-machine setup: you might want to sanity-check the connection, or see what models are on the server, without also spending a GPU cycle running an actual upscale. It's a cheap diagnostic step you can drop anywhere in a graph while you're getting the server side working.