ComfyUI with ROCm on Windows 11: Z Image Turbo setup

This English summary was reviewed against the available Arabic captions. This tutorial walks through ComfyUI installation on Windows 11 with ROCm for AMD hardware, then uses Z Image Turbo to generate an image locally. The presenter identifies Radeon 7000 series GPUs with RDNA 3, 9000 series GPUs with RDNA 4, and Ryzen AI Max with RDNA 3.5 as supported hardware in the repository instructions shown.

The setup starts with Git for Windows and Python 3.13. The presenter explains how to select Python 3.13 when multiple Python versions are installed, clones the ComfyUI repository, and creates an activated virtual environment. The AMD section of the repository supplies the installation command for torch, torchaudio and torchvision through ROCm. Selecting a specific GFX target instead of all devices can reduce the packages installed. ComfyUI's requirements follow before launching main.py.

In the browser interface, the Z Image Turbo text-to-image template initially reports missing models. The tutorial places the Z Image Turbo BF16 file, AE VAE file and Qwen 3 four-billion-parameter text encoder in their respective model folders, then refreshes ComfyUI.

The example generates a 2D cartoon of a swan on a bicycle. The presenter uses an AMD GPU with 24 GB of VRAM and the high-VRAM option, reporting about one iteration per second. That result describes this demonstration; the presenter also says the Windows metrics are not fully accurate.