ComfyUI: understand, install, troubleshoot

Installing ComfyUI: which installation for which use

The official methods (Desktop, portable, manual), what you need to know before installing, and the layer-by-layer reading that lets you diagnose instead of reinstalling everything.

Installing ComfyUI starts with choosing the method according to the system, the GPU, where the models are stored and the level of control you want. The official methods, Desktop, portable on Windows and manual installation, cover Windows, macOS and Linux; for a production workstation, the manual installation keeps control over Python, PyTorch, CUDA, the dependencies and the model paths. At Axiris, we read an installation layer by layer, from the driver to the model, to know which one breaks; and for those who do not want to install, a machine in Switzerland is delivered ready.

Choose before you install

Installing ComfyUI is not just downloading a program. The right choice depends on the system, the GPU, how the models will be stored, and the level of control you want to keep over the Python environment.

ComfyUI offers several official approaches: a Desktop version, a portable version on Windows and a manual installation. The official documentation covers Windows, macOS and Linux.

At Axiris, we distinguish two needs.

The simple installation. To discover ComfyUI or quickly have a working environment, the methods ComfyUI offers directly are enough. You learn the software, not its installation.

The controlled installation. For a production workstation, a dedicated machine or an environment that depends on particular libraries, the manual installation keeps control over Python, PyTorch, CUDA, the dependencies and their versions, the launch parameters, the extensions and the paths to the models. This control matters as soon as you work with recent models or with extensions that have specific dependencies: that is where simple installations get stuck.

What you need to know before you start

Before the first command, identify:

  • the operating system;
  • the GPU and its amount of VRAM;
  • the NVIDIA driver version, where applicable;
  • the available disk space;
  • how the models will be organised.

Models quickly add up to tens or hundreds of gigabytes. If ComfyUI is to become a production tool, the storage structure is thought through at the start, not afterwards: see models in ComfyUI and GPU, VRAM and AI image models.

The official methods

The official documentation details each installation:

The Axiris approach: understanding the environment layer by layer

We prefer to understand the environment rather than apply a series of commands. When an installation breaks, what helps is knowing which layer is the problem:

driver → CUDA → PyTorch → Python → dependencies → ComfyUI → extension → model

Each layer rests on the previous one. An error is read by going back up this chain, and it almost always has a single address. This reading lets you diagnose instead of reinstalling the whole machine: the method is detailed in Troubleshooting ComfyUI.

The other way: not installing

Installing and running a production workstation takes time and hardware. For a business, Axiris offers a machine in Switzerland, rented by the hour and delivered ready: the system, the drivers, ComfyUI and Nova are installed, you log in and you work. For a creative, you start on the studio's machine, without investing.

A question about your installation?

The Synnexia ComfyUI AI can be asked about an installation, an error or a particular configuration.

To set up your production environment with Axiris: Contact request: a first conversation, free and with no commitment. Every practice is different; it is by listening to you that we propose what suits you.