ComfyUI-ComfyLink

ComfyUI-ComfyLink — run your ComfyUI from your phone, from anywhere

ComfyLink lets you remotely control your own ComfyUI from a mobile app (iOS & Android) — from anywhere, with a real prompt manager. Run your saved workflows from your phone, tweak the prompt and parameters, tap generate, and the results come back to your device — while the generation runs on your own PC/GPU at home.

If you've ever wanted a ComfyUI mobile app, remote access to ComfyUI, or a way to start a generation from your phone and get the result on your phone, that's exactly what this is. This repo is the ComfyUI custom node; it pairs your local ComfyUI with the ComfyLink app.

• 📱 App: iOS (App Store) & Android (Google Play) — search "ComfyLink"
• 🖥️ Your PC does the work; the app is just the remote control.
• 🔒 Outbound-only, pair-once — no port forwarding, no VPN, no cloud GPU, and your PC is never exposed to the internet. Works on WiFi or 4G.
• 🧩 A real prompt manager on your phone — not just a text box.

📖 New here? Full setup & usage guide → https://comfylink.app/guide

How is this different from other ComfyUI mobile tools?

Most ComfyUI mobile tools are just a remote screen: they need your phone on the same LAN (or a port-forward / VPN / cloud box to reach your PC), and prompts are a plain text box — you retype or paste every time. ComfyLink is different on three fronts:

1. Works from anywhere, safely. The plugin is outbound-only — your PC connects out to a lightweight relay, so it works behind home NAT with no port forwarding, no VPN, and nothing exposed to the public internet. On WiFi or on 4G, same thing.
2. A real prompt manager, on your phone. Not a text box — organize prompts into presets and categories, star your favorite terms into a reusable library, tune per-term weights, and drop a whole preset into any workflow with a tap. Build your prompt library once, reuse it everywhere.
3. It knows what's installed on your machine. On request, the plugin reports your LoRA and checkpoint inventory — including the trigger words model authors embed in their own files — so the app can offer real models and real trigger words instead of you typing filenames from memory.

How it works

• Outbound only — no port forwarding; works behind home NAT.
• Pair with a one-time code from the app. The PC never stores your account password. More than one account can pair with the same ComfyUI.
• Results are delivered to your phone — images, and also animations and videos (VHS_VideoCombine, SaveVideo, animated WebP/GIF). Outputs are staged only briefly on the way to your device and are auto-cleared after ~2 days — they don't pile up on the relay.
• Keep using ComfyUI normally. App jobs and anything you run locally share the same queue (one GPU), but each job's outputs stay with that job — your local generations never get mixed into the app. Cancelling from the app is surgical: a job still queued is removed by name, and only a job that is actually running gets interrupted — the plugin never fires a global interrupt that could kill the generation you started yourself.

Requirements

• A working ComfyUI install, recent enough to have the new frontend sidebar (the ComfyLink panel registers as a sidebar tab — without it, the panel won't appear and there is no other way to pair).
• Python 3.10+.
• The ComfyLink app, signed in.

New to ComfyUI? See the official docs to install it and learn the basics:

• Documentation: https://docs.comfy.org
• ComfyUI repository: https://github.com/comfyanonymous/ComfyUI
• Example workflows: https://comfyanonymous.github.io/ComfyUI_examples/

Install

With ComfyUI-Manager (easiest). Open Manager → Custom Nodes Manager, search for ComfyLink, click Install, then restart ComfyUI. Dependencies are installed for you.

Or clone it manually into your ComfyUI custom_nodes folder and restart ComfyUI:

cd ComfyUI/custom_nodes
git clone https://github.com/huijiutian/ComfyUI-ComfyLink.git

Cloning tracks main, which is where development happens — you get changes before they're cut into a release. Installing through Manager keeps you on published versions.

Dependencies

Both are normally already in your ComfyUI environment, and both are declared in requirements.txt with loose >= bounds so pip won't downgrade what ComfyUI ships:

• aiohttp — required. Used to talk to the relay.
• Pillow — optional but recommended. Only used for the app's "convert outputs to WebP" option, and for writing the prompt into the image (see Round-trip below). Without a working Pillow the job still succeeds — you just get the original PNG instead of WebP. If Pillow is present but too old to write XMP, the panel says so.

Pair

1. After restarting ComfyUI, open the ComfyLink panel in the sidebar. It shows the connection status.
2. In the ComfyLink app, tap Pair ComfyUI to get a one-time code.
3. Paste the code into the panel, optionally give this machine a device name, and click Pair.

That's it — the panel turns Online and your machine shows up in the app, ready to generate. No config files to edit.

The pairing form stays available afterwards: one ComfyUI can serve several accounts (yourself on two accounts, or a machine shared with someone you trust). Each paired account is listed in the panel with its own Unpair button, so you can revoke one without touching the others.

Choose which workflows the app sees

Once paired, the panel shows a Manage workflows list of your saved ComfyUI workflows. Tick the ones you want available on your phone, tick which paired accounts to sync them to, then click Upload / update selected — they're converted and pushed to the app. Each row shows when it was last uploaded, and is tagged changed if you've edited the workflow since; re-upload to refresh it. Refresh list re-reads your workflows from ComfyUI. Only the workflows you select appear in the app, and uploading never deletes anything on the app side.

On the phone, they don't appear on their own: open the Workflows tab, tap sync on this ComfyUI's row, and pick which ones to import. Full walkthrough: https://comfylink.app/guide

Let the app see your models

The app can ask this machine two things, both only when you tap refresh in the app — there is no background scanning:

• Your model inventory. The plugin lists your LoRAs and checkpoints and reads each file's embedded header: for LoRAs the trigger phrase the author put in the file plus the most frequent training tags; for checkpoints the model family (SDXL / Flux, eps vs v-pred, training resolution, and so on) so the app knows which prompt dialect fits. No model file's content is ever read — one stat plus the safetensors header, which is why a few hundred models scan almost instantly. Nothing is hashed and nothing is uploaded to a third party.
• A fresh node/model snapshot (object_info). This is where the app's model dropdowns come from, and it used to be captured only once at startup — so installing or deleting a model on your PC would leave the app choosing from a stale list and the job would fail. Now you can refresh it from the app, and the plugin reports back as soon as it's done.

Related: when a workflow references a model this machine no longer has, the plugin parses ComfyUI's validation error and tells the app which model is missing by name, instead of a wall of JSON.

Prompts land in your ComfyUI log

Right before a job is submitted, the plugin prints the final prompt text to your own ComfyUI console — per text input, labelled positive or negative, with the node it belongs to. The roles are traced back from the sampler along the graph, so prompts behind FluxGuidance, ConditioningCombine or text-concat nodes are still located correctly.

This makes the phone-side prompt manager auditable: what you assembled on your phone is printed on your machine, before the image exists. It is logged locally only and never sent to the relay.

Round-trip: your own images back into a preset

When WebP output is enabled, the plugin also writes the prompt into the image's XMP as plain, directly readable fields (comfylink:positive / comfylink:negative, plus a comfylink:generator marker) — alongside the full workflow JSON. Drop one of your own images back into the app and "build a preset from this image" reads the prompt straight off it, with no guessing about node structure. If the prompt can't be determined, nothing is written rather than something wrong.

Status & control

The panel shows whether this PC is Not paired / Connecting / Online (plus · generating while a job is on the GPU), the machine name, the number of detected nodes, the list of paired accounts, and the running version and commit. It also surfaces:

• a soft update hint when a newer plugin version is published;
• a red banner if this plugin is too old for the relay to accept jobs;
• an amber warning if Pillow on this machine can't write the prompt into WebP output.

You can Unpair any account here anytime; you can also unpair from the app. Either way, access stops immediately.

Inert until paired. Before you pair, the plugin takes no jobs and uploads nothing. (Opening the panel does make one small outbound request — a version check against the relay, cached for 30 minutes. It carries no credentials and none of your data.)

Staying up to date

If you installed through ComfyUI-Manager, update it there. If you cloned it, pull and restart ComfyUI:

cd ComfyUI/custom_nodes/ComfyUI-ComfyLink
git pull

The sidebar panel shows the running version and commit, and hints when a newer version is available, so you can confirm you're on the latest.

FAQ

Can I run ComfyUI from my phone? Yes. ComfyLink is a mobile app (iOS & Android) that runs your existing ComfyUI workflows remotely. You install this custom node on your PC's ComfyUI, pair it once with the app, and then trigger generations from your phone — the results come back to your device.

Is there a ComfyUI mobile app? ComfyLink is a mobile companion for your own ComfyUI. It doesn't generate images in the cloud — your PC/GPU does the work, and the app is the remote control. You pick a workflow, adjust the prompt/parameters, generate, and view results on your phone.

How do I access ComfyUI remotely without port forwarding or a VPN? The node is outbound-only: your PC connects out to a small relay, so it works behind home NAT with no port forwarding, no reverse proxy, and no VPN. You pair with a one-time code; nothing is exposed to the public internet.

Can I manage prompts on my phone, or is it just a text box? It's a full prompt manager, not a plain text box. Organize prompts into presets and categories, star favorite terms into a reusable library, tune per-term weights, and drop a whole preset into any workflow with a tap — all on your phone. Most mobile ComfyUI tools only let you retype prompts in a text field; ComfyLink lets you build a prompt library once and reuse it everywhere.

Does it run generation on someone else's servers / a cloud GPU? No. All image generation runs on your own hardware. Our relay only passes small job messages and briefly stages the output so it can reach your phone (auto-cleared after ~2 days). No cloud GPU, no images kept long-term on our side.

Can I keep using ComfyUI on my PC while using the app? Yes. App jobs and anything you run locally share the same ComfyUI queue (one GPU), but each job's outputs stay with that job — your local generations are never mixed into the app's gallery, and cancelling an app job never interrupts one of yours.

Which workflows show up in the app? Only the ones you pick. After pairing, use Manage workflows in the panel to select which saved workflows are pushed to the app, and to which paired accounts.

Does it scan my model folders? What exactly gets sent? Only when you ask for it from the app — there is no background scanning. The plugin sends filenames, sizes and the metadata embedded in each safetensors header (trigger phrase, training tags, model family). It never reads a model file's content and never hashes it.

Can more than one person / account use the same ComfyUI? Yes. Pair each account from the panel; every account gets its own entry and its own Unpair button. They share the machine's single GPU queue.

Does it do video, or only images? Video and animation too — outputs from VHS_VideoCombine, native SaveVideo, and animated WebP/GIF are delivered like any other result. Video is never re-encoded to WebP.

Is it free? What are the limits? The core remote-control features work on the free tier. Paid tiers (Plus/Pro) raise usage limits (prompt presets, per-generation size, staging capacity). See the app for the current numbers.

iPhone or Android? Both — the app is on the App Store (iOS) and Google Play (Android). Search "ComfyLink".

Keywords

ComfyUI mobile app · run ComfyUI from phone · remote control ComfyUI · ComfyUI remote access · ComfyUI iOS app · ComfyUI Android app · control ComfyUI from iPhone · trigger ComfyUI workflow from phone · self-hosted Stable Diffusion remote

License

MIT