Add proper cloud comfy endpoint support (e.g. runpod)

[saniul1]

• right now its confusing how to set up cloud endpoint.
• custom nodes install in remote machine is confusing (make manual install easy)
• if user choose to use the app only to connect to remote comfy instance. let us skip the local comfy installation

[Mooshieblob1]

I checked this against the work currently in my local branch, and this issue is not being addressed there yet.

The changes I have in progress are more about browser mode, GPU workers, style transfer support, and related infrastructure. They do not yet provide the clearer remote cloud endpoint setup, simpler remote custom node flow, or remote-only path that would let users skip local Comfy installation.

I am leaving this open.

[Mooshieblob1]

Added a remote-only setup path in the current branch. First-run setup can now be completed without a local ComfyUI install, the wizard validates the remote endpoint before finishing, and it shows a manual checklist when the remote server is reachable but missing MooshieUI custom nodes.

[Mooshieblob1]

I just pushed the current fix for this. It should be in the next release. Please keep this ticket open until that release is out and you have had a chance to confirm it behaves as intended.

[saniul1]

Could not connect to remote server. not clear how to install MooshieUI custom nodes on remote server.

[Mooshieblob1]

Thanks for raising this. Here is what we have done and how remote/cloud ComfyUI is meant to work.

Skip local ComfyUI install

The setup wizard now has a Remote server option on first launch. Choose that, paste your cloud ComfyUI URL, and click Validate remote server. MooshieUI will not install Python, PyTorch, or ComfyUI locally.

Cloud endpoints (RunPod, etc.)

In Settings → Connection, set Server Mode to Remote and paste the full public ComfyUI URL (for example https://xxxxx-8188.proxy.runpod.net). The README now includes a Remote / cloud ComfyUI section with deploy and connect steps.

Custom nodes on the remote server

MooshieUI custom nodes must be installed on the remote server via the MooshieUI server build (Docker). The server image includes ComfyUI and all required MooshieUI nodes, plus the ControlNet dependencies our presets need. Manual node copying is not supported.

Deploy on your cloud GPU with:

docker compose up -d --build

or pull ghcr.io/mooshieblob1/mooshieui:latest, expose ComfyUI on port 8188 (or use the MooshieUI web UI on port 3200 directly), then connect the desktop app to that endpoint.

If validation fails because nodes are missing, redeploy or update the server build on the cloud instance and fully restart ComfyUI there before retrying.

Let me know if you hit a specific RunPod or cloud template that still feels unclear and we can document it.

[Mooshieblob1]

Quick follow up: v1.4.2 is out now with the remote/cloud ComfyUI work described above. Please update when you can and let me know if your RunPod or cloud setup still needs anything clarified.

[saniul1]

i was able to connect to my remote server. I copied the custome_nodes from my local installion to the remote machine. and it worked. it would be better to know which files/folder should be copied to remote. but the main problem is i was not able to generated any image as its want to download anima model which i have already downloaded and testen in my remote machine.please fix the issuie of remote model detection. thank you.

[saniul1]

The app needs proper testing with remote instances. It seems a lot of changes needed to properly work with remote comfy installation. It is currently focuses on local setup. there is lot of edge cases i'v found that should be easier for you to fix if you use a remote instance. rather than me trying to list all of them.

[Mooshieblob1]

Thanks for sticking with this and for the detailed follow up.

v1.4.2 covered the connection side: remote setup in the wizard, skipping the local ComfyUI install, and docs for the Docker server path. You got connected, which was the main goal for that release.

The model detection problem you hit is real. A lot of the app still checks your local disk for models (hash scans, download prompts, model manager stuff) even when you are in remote mode. If Anima is already on the remote server, the app should trust what ComfyUI reports over the API instead of asking you to download it again. That needs a proper pass through the codebase, not a quick patch.

Same for the broader remote edge cases you mentioned. Remote mode works for connect and generate, but a bunch of features still assume a local ComfyUI install. Fixing that properly means testing against a real remote instance and updating model handling, downloads, and probably some of the setup flows. That is bigger than what fits in a single issue fix.

I am going to leave this open as a tracking item but treat full remote parity as future work rather than something I am actively chipping away at right now. When I pick it up again, remote model detection is probably the first thing to tackle.

If you have a specific RunPod or cloud template that still feels unclear in the docs, feel free to drop it here and I can add a note to the README even before the bigger remote work.