Be Confident: How to Sandbox ComfyUI to Protect Your OS from Malware

Using WSL 2 and Ubuntu, you can confidently run 3rd-party workflows locally

If you’re back for Part 2 of this ComfyUI series, that means you took the recommendation from Part 1 seriously and considered whether ComfyUI is the right tool for you. And here you are, wanting to learn more.

That’s exciting!

In this Insight, I’ll guide you through the initial steps of the ComfyUI installation process, ensuring it is completely isolated from the host operating system. Such an installation will make ComfyUI very safe and secure.

Why is this important? ComfyUI is very hackable, which, counter-intuitively, is great because anyone with Python skills can write custom nodes. Unfortunately, sometimes third-party ComfyUI workflows that are shared in online communities contain nodes infected with malware.

The foundation we lay down in this and the following several tutorials will be helpful should you venture beyond ComfyUI to explore other local AI tools. We’re building something that will come in handy if you choose to explore things beyond Comfy UI – it’s a great foundation for any AI-related tasks you want to perform on your machine.

“The problem with Comfy UI is that while the software is perfectly safe…it allows third-party users to write their own nodes and their own workflows…in Python. And they can create a lot of damage on your machine.”

Igor Riđanović, Finisher / Colorist

There are generally three ways you can get ComfyUI running.

One, you can go to comfy.org and download and install a desktop app for Mac OS or Windows. That’s a fast way to get up and running, but the problem is you’re running ComfyUI on your local machine unprotected. You’re probably OK if you stick to the ComfyUI core nodes, but you can’t really download anything else. And that’s too bad because the real power of ComfyUI comes from all these third-party workflows – but they could be dangerous.

Two, you can use an online cloud-based service. The problem with that is it won’t be free. So you’re back in the same boat as if you’re using Sora, Runway, Google VO3, or whatever. You’ll be spending money. And the second problem is even greater than that – those cloud service providers can peek at your files and everything that you’re doing in ComfyUI. I’m not sure about you, but that is a non-starter.

Three, you can install ComfyUI from the open-source that’s hosted on GitHub. That’s the way to do it if you’re running Linux. However, you can also do it on Windows and macOS if you want to get the latest version.

Instead of installing directly, we’ll set up a virtual machine and a container for ComfyUI, effectively hiding our host operating system. If ComfyUI were to load a malware-infected node, the worst that such a node could do would be to damage the container. This is a very small price to pay because containers are easy to rebuild.

Important note: In this Insight, we’ll cover the setup of WSL 2 (Windows Subsystem for Linux) and the installation of a Linux Ubuntu virtual machine. If you already have Linux, you can skip this episode. Macs do not support Nvidia cards, so setting up macOS is outside the scope of this ComfyUI series. It’s possible to run ComfyUI on a Mac with a small performance decrease compared to machines with an Nvidia GPU, but we won’t be covering that installation here.

Key takeaways from this Insight

By the end of this Insight, you should understand how to:

Related Mixing Light Insights

Mentioned and External Resources