ComfyUI integration

ComfyUI integration

Use your own ComfyUI workflows inside Magnific. They keep running in ComfyUI on your own machine — your hardware, your models — while Magnific gives them a clean interface and lets you combine them with other tools in a Spaces pipeline.

Closed alpha. This is early, invite-only software. Expect rough edges and breaking changes between versions; features may move or disappear. Don't rely on it for production or business-critical work yet. Support is limited, and your feedback at this stage is especially valuable.

The three parts, and how your workflow becomes a Flow

Three pieces work together:

• Your ComfyUI — runs on your own machine or server, exactly as it does today. The workflow executes here.

• The Magnific plugin — a small add-on you install inside ComfyUI. It's the bridge: you mark which inputs Magnific can fill in and which outputs come back to it. Everything else stays untouched.

• Magnific Spaces — where you and your team run the workflow and see the results.

Once you register a workflow, your whole ComfyUI workflow becomes a single node in Magnific called a Flow. It takes the inputs you defined, runs the full workflow on your machine, and returns the output — all from one node on the canvas.

When someone runs a Flow, only the inputs travel to your machine and only the outputs travel back. The workflow itself never leaves ComfyUI. Magnific never hosts it or sees what happens inside.

Your machine only makes outbound calls

Your ComfyUI machine doesn't need a public IP, open ports, or a tunnel. It only makes outbound calls to Magnific.

The plugin checks in with Magnific over HTTPS and waits for a job. When one arrives, ComfyUI runs it locally and the plugin uploads the result back out the same way. Magnific never initiates a connection to your machine. For the exact hosts and ports your IT or security team needs, see the section with Network & ports below.

What travels to Magnific, and what's stored

The workflow itself never leaves ComfyUI — Magnific never receives it or sees what happens inside.

What does travel: the inputs you send a Flow and the outputs it returns. These pass through Magnific's media CDN (*.cdnpk.net) so they can be displayed and used in Spaces, the same way any other Magnific creation is. They are not used to train models, and they are subject to your organization's standard Magnific data-retention terms.

Your sign-in is handled by a short-lived access token that refreshes automatically, so no long-lived secret is written to disk. It's stored locally on your own machine, never on Magnific.

What you need before you start

Make sure you have everything below on the machine where ComfyUI runs.

Install the plugin in ComfyUI

Follow these steps in order, on the machine where ComfyUI runs.

1. Locate your ComfyUI folder. This is the directory that contains main.py and a custom_nodes/ subfolder; every step below is relative to it. In the Desktop app, the quickest way to open it: click your instance name in the top bar to open the instance panel, go to the Terminal tab, and run open . (macOS) or start . (Windows). That opens the ComfyUI folder in your file browser.

2. Download the release archive containing the plugin. To receive the download link, contact your organization admin.

3. Extract it into custom_nodes/. When you're done, you must have exactly this path — the folder name matters: ComfyUI/custom_nodes/comfy_remote/. If you instead see comfy_remote/comfy_remote/ or a versioned folder name, move or rename it so the plugin files sit directly inside comfy_remote/.

4. Install soundfile, the plugin's one Python dependency, into the same Python that runs ComfyUI. It's only needed for audio inputs. The exact command depends on how you installed ComfyUI — see "Install soundfile" below. If you install the plugin through ComfyUI-Manager instead, it does this for you.

5. (Optional) Install ffmpeg if you'll use audio. It's a reliable fallback decoder.

6. Restart ComfyUI fully. Stop the process and start it again — a browser refresh is not enough.

When ComfyUI restarts, you should see a Magnific tab in the left sidebar, plus the Remote Inputs and Remote Output nodes in the add-node menu.

If the Magnific tab doesn't appear, the most common cause is that ComfyUI wasn't fully restarted, or that soundfile was installed against a different Python than the one ComfyUI uses. Re-run the soundfile install below against the correct interpreter.

Install soundfile

soundfile must go into the same Python environment that runs ComfyUI — not your system Python. Use whichever method matches your install.

Desktop app (simplest). Open the Terminal tab in the ComfyUI Desktop app. Its prompt starts with (.venv) and opens in your ComfyUI folder — that means ComfyUI's own Python is already active, so there's nothing to locate. Just run:

pip install -r custom_nodes/comfy_remote/requirements.txt

Tip: typing open . (macOS) or start . (Windows) in that terminal opens the current folder in your file browser — handy for dropping the plugin into custom_nodes/ in step 3.

Portable build (Windows only). From the portable root folder:

python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\comfy_remote\requirements.txt

Manual install (git + venv). Activate the environment you created, then install:

# macOS / Linux
source venv/bin/activate
pip install -r ComfyUI/custom_nodes/comfy_remote/requirements.txt

# Windows
venv\Scripts\activate
pip install -r ComfyUI\custom_nodes\comfy_remote\requirements.txt

Not sure which Python ComfyUI uses? When it launches, it prints a line like Python executable: /path/to/python in the startup log. Use exactly that path.

What your network needs to allow

You don't need to open or forward any inbound ports, and you don't need a public IP. The plugin only makes outbound connections, and Magnific never connects to your machine.

If your company uses an egress firewall or HTTP proxy, allowlist outbound port 443 to these hosts:

• mcp.magnific.com

• auth.magnific.com

• .cdnpk.net (media upload and download)

That's the one piece enterprises need to configure. Everything else is local.

Register a workflow as a Flow and run it

Step 1 - Sign in (once per machine). In ComfyUI, open the left sidebar → Magnific tab → Authenticate.

Step 2 - Build your workflow. Create a normal ComfyUI workflow that produces what you want.

Step 3 - Add a Remote Inputs node. In its panel, add one row per input your workflow needs. Each row has a name and a type (string, int, float, image, audio, or video). Wire each output socket into your graph.

Step 4 - Add a Remote Output node (Image, Audio, or Video) for every result you want sent back.

Step 5 - Copy for Magnific Spaces. Open the Magnific sidebar and click Copy for Magnific Spaces.

Step 6 - Paste into Magnific Spaces. Paste the node onto a Space and set the inputs. This opens a wizard that creates the Flow. Only an authorized account can paste a workflow into Spaces.

Step 7 - Wire it up and run it. Your Flow appears on the canvas as a ComfyUI flow node. Connect an input to each field you defined, and an output node to catch the result. For a text-to-image workflow with one prompt input, that's: a Text node (horse) → the ComfyUI flow node → a List node for the images. Press Run flow. You can also run it as a standalone form, or share it with your team (see "Share a Flow" below).

The input name is the contract. A field named cfg on the canvas fills the input named cfg in your workflow. And Spaces always runs the version you last saved, not whatever is open in your editor — after editing, save again.

Rules for inputs and outputs

• Default values. By default, remote inputs use the connected node's default values. Some of those only change in the ComfyUI frontend (for example, seed), and some are tensors or latents that can't be passed remotely. Valid input types are string, int, float, image, audio, and video.

• Supported media. Audio inputs and outputs are MP3. Image and video inputs arrive as decoded Magnific creations — wire them straight into your graph.

• Booleans. There's no boolean type yet. Nodes like KJNodes' bool typically use binary values (0 / 1) for false / true.

• Output naming. You can rename remote outputs, and we recommend doing so when you have more than one. A valid remote output is currently required so Spaces knows the run has finished. (A "signal" output type for void returns is in progress.)

What a run costs in credits

The generation runs on your own ComfyUI machine, on your own hardware, so Magnific isn't charging you for the compute. A ComfyUI Flow run therefore costs little or nothing in credits — the node shows an estimate before you run, and for a typical run it's close to zero.

Change a Flow's inputs and re-sync it

If you need to add, remove, or change an input, re-sync the Flow in Spaces. After saving your changes in ComfyUI:

1. Re-copy the workflow for Magnific Spaces (left sidebar → Copy for Magnific Spaces).

2. In Magnific Spaces, go to Flows.

3. Open the list and select your Flow.

4. Click Edit and paste your ComfyUI workflow configuration. The wizard autodetects the changes and offers to update the Flow in Spaces.

Upgrade the plugin to a new version

Upgrading is a folder swap. Your settings live elsewhere, so nothing is lost.

1. Download the newer archive, e.g. comfy_remote-vX.Y.Z.zip.

2. Remove the old version: delete the existing ComfyUI/custom_nodes/comfy_remote/ folder. This is safe — your sign-in and saved workflows aren't stored there (see step 6).

3. Extract the new zip into custom_nodes/ so you again have exactly ComfyUI/custom_nodes/comfy_remote/.

4. Re-install soundfile in case its version changed, using the same command as a fresh install (see "Install soundfile" above).

5. Restart ComfyUI fully.

6. Confirm the version. Your authentication token and armed workflows live outside the plugin folder — on macOS and Linux in ~/.config/comfyui-remote/, and on Windows under your user profile — so after an upgrade you don't need to sign in again or re-arm your workflows. The running build is reported to Magnific — check it in the Magnific sidebar panel.

If something behaves oddly right after an upgrade, the cause is usually the same as a fresh install: ComfyUI wasn't fully restarted, or soundfile was installed against the wrong Python. Re-run the soundfile install with the correct interpreter.

Share a Flow: private, project, or organization

1. In Magnific, open Spaces.

2. Go to your saved Flows.

3. Choose the Flow's visibility:

   • Private — only you.

   • Project — members of a shared project.

   • Organization — your whole team.

   You can also click Share with team to share via a link.

Once shared, teammates find the Flow in their Home under My flows, or in the Library.

During the alpha, keep the host machine online

Shared runs still execute on the ComfyUI instance that armed the workflow. That machine must be running and signed in to Magnific. If it's offline, your teammates' runs will wait and eventually fail.

Runs are processed through that host's ComfyUI queue, in the order they arrive. If several teammates trigger a Flow at once, their runs line up and run one after another — so a busy or slow host means longer waits for everyone. Share widely used Flows from a host that can keep up.What to keep in mind during the alpha

The host ComfyUI must stay online. Shared runs still execute on the ComfyUI instance that armed the workflow. That machine must be running and signed in to Magnific. If it's offline, your teammates' runs will wait and eventually fail.

Troubleshooting and common questions

• It runs, but nothing happens / the Spaces job never completes. Make sure ComfyUI is running and signed in (the Magnific tab shows as connected), and that you saved the workflow after your last edit. The background worker only runs while ComfyUI is open and authenticated.

• My change in ComfyUI didn't take effect in Spaces. Spaces runs the last saved version. Save the workflow again (⌘S) after editing, then re-copy it for Spaces.

• Audio input comes back silent. Audio creations are MP3. Make sure soundfile is installed (pip install -r requirements.txt); installing ffmpeg as well gives a reliable fallback. Then run again.

• Do I need to expose a port or a public IP? No. The plugin only makes outbound connections to mcp.magnific.com. See "What your network needs to allow" above.

• Where are my credentials stored? On macOS and Linux, in ~/.config/comfyui-remote/; on Windows, under your user profile. The access token is short-lived and refreshes automatically, so no long-lived secret is written to disk.

• I can't sign in. Your account probably isn't provisioned for the alpha yet. Reach out to your Magnif