Simplify ComfyUI on Cloud GPUs

If you've ever spun up a cloud GPU instance on Vast.ai or RunPod to run ComfyUI, you know the drill: wait for the instance to start, manually download models, install custom nodes, update ComfyUI to the latest version, restart everything, and hope you didn't miss a step.

It's tedious, time-consuming, and since you're paying by the minute, it's costing you money.

I built a simple (and free) tool to fix this 👉 deploy.promptingpixels.com

What It Does

The deploy script generator creates a one-liner bash command that handles all your setup automatically:

You configure what you need once, copy the command, paste it into your cloud instance's terminal, and walk away.

When it's done, your environment is ready to go with the exact models, custom nodes, and ComfyUI version you specified.

How to Use It

Let me walk you through a typical setup.

Start by launching your cloud GPU instance. I'll use Vast.ai with the official ComfyUI template as an example.

Once it's running, open the Jupyter Terminal—you'll have a basic bash prompt ready to go:

1. Head over to deploy.promptingpixels.com and click "Choose App." You'll see ComfyUI selected by default along with the latest version, which is important since the version pre-installed on most cloud templates tends to lag behind.
2. Next, click "Add Models" to configure which models you want downloaded. The tool pulls from both Hugging Face and Civitai, so most popular models are available. For a quick test, you could grab something like DreamShaper 8 from Hugging Face. The key thing happening here is that the script maps each model file to the correct directory structure within ComfyUI—checkpoints go to the checkpoints folder, LoRAs to the LoRA folder, and so on.
3. Custom nodes work the same way. Click "Custom Nodes" and search for whatever you need. The tool pulls from the ComfyUI registry, so popular nodes like Impact Pack are easy to find and add.

Once you've configured everything, you'll see a one-liner script at the bottom of the page. Copy it, paste it into your cloud instance terminal, add your Hugging Face token if you're downloading gated models, and hit enter:

The script handles the rest.

After it finishes, launch ComfyUI from your instance portal. Your models will be there, your custom nodes will be installed (you may need to restart ComfyUI once for nodes to appear in the menu), and you'll be running the latest version.

A Few Things Worth Noting

If you want to see exactly what the script does, click "Full Script" to view the underlying bash commands. You can copy the full script and modify it however you like:

The tool supports multiple providers. Vast.ai and RunPod have slightly different directory structures, so make sure you select the correct provider before copying your script. You can even use it with Windows, Mac OS, and Linux operating systems by simply defining where you have ComfyUI installed on your machine.

I've also added preset configurations based on setups from my previous tutorials. If you watched my Qwen Image Edit with ControlNet video and want to recreate that exact environment, you can load that preset and it'll pull in all the relevant models, custom nodes, and the specific ComfyUI version I was using.

Questions or Feature Requests?

If you run into issues or have ideas for improvements, drop a comment on the video or email me directly at deploy@promptingpixels.com. I'm actively maintaining this tool and happy to add features that would help your workflow.