Setting Up ComfyUI for AI Image Generation on Windows with AMD RDNA 4 Using ROCm 7.1.1

ROCm 7.1.1 is a major turning point for AMD Radeon users who want to run AI workloads on Windows. For the first time, RDNA 4 GPUs can tap into their full performance for local AI tasks without being pushed into Linux or relying on workarounds like WSL or compatibility layers. The big headline here is PyTorch support on Windows through ROCm—meaning popular workflows like local AI image generation finally feel “native” on Radeon’s latest architecture. (Worth noting: TensorFlow, Jax, and ONNX support for RDNA 4 still remains Linux-only for now.)

If you’ve ever tried setting up ComfyUI on an AMD GPU and felt like the process dragged on forever, this is the update you’ve been waiting for. What used to take days—or even weeks—of troubleshooting can now be done in a clean, straightforward setup that gets you generating images quickly with far fewer headaches.

Start with the right driver and tools

To enable ROCm 7.1.1 PyTorch support on Windows, begin by installing AMD’s 25.20.0.17 driver. It’s designed to work with RDNA 3, RDNA 3.5, and RDNA 4 graphics cards, although compatibility lists can sometimes look oddly selective.

Next, install two essentials:
1) Miniconda (so you can isolate dependencies cleanly)
2) Git (so you can download ComfyUI easily)

Using a dedicated conda environment is strongly recommended here. It keeps your ROCm + PyTorch setup from colliding with other Python projects on your PC.

Create a new conda environment (Python 3.12 matters)

Open “Anaconda Prompt” and create a new environment like this:

conda create –name insertnamehere python=3.12

Python 3.12 is important because ROCm 7.1.1 PyTorch wheels are built for it. If you use a newer Python version, you’ll likely hit compatibility issues.

Activate your environment:

conda activate insertnamehere

Install PyTorch for ROCm 7.1.1 (Windows)

With the conda environment active, install the ROCm SDK components first (these are required for the ROCm-enabled PyTorch setup on Windows). Run:

pip install –no-cache-dir ^
https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm_sdk_core-0.1.dev0-py3-none-win_amd64.whl ^
https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm_sdk_devel-0.1.dev0-py3-none-win_amd64.whl ^
https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm_sdk_libraries_custom-0.1.dev0-py3-none-win_amd64.whl ^
https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm-0.1.dev0.tar.gz

Then install PyTorch and key components (TorchAudio and TorchVision) built for ROCm 7.1.1 on Windows:

pip install –no-cache-dir ^
https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torch-2.9.0+rocmsdk20251116-cp312-cp312-win_amd64.whl ^
https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torchaudio-2.9.0+rocmsdk20251116-cp312-cp312-win_amd64.whl ^
https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torchvision-0.24.0+rocmsdk20251116-cp312-cp312-win_amd64.whl

After installation, confirm PyTorch can see your GPU. Run:

python -c “import torch; print(f’device name [0]:’, torch.cuda.get_device_name(0))”

If everything is correct, it should output your GPU’s name. At this point, your environment is ROCm-enabled on Windows and ready for accelerated AI workloads.

Install ComfyUI on Windows (RDNA 4 ready)

Choose where you want ComfyUI installed (for example, your Desktop), navigate there in Anaconda Prompt, and clone the project:

cd Desktop
git clone https://github.com/comfyanonymous/ComfyUI.git

Go into the ComfyUI folder:

cd ComfyUI

Install required dependencies:

pip install -r requirements.txt

Launch ComfyUI:

python main.py

Once it starts, you’ll see a local web server address. Hold Ctrl and click the link to open the ComfyUI interface in your browser.

Set up a model and generate your first local image

Inside the ComfyUI interface:
1) Open Templates on the left
2) Search for “SDXL Turbo”
3) Download the model when prompted
4) Move the downloaded model file into:
ComfyUI > Models > Checkpoints

Now you’re ready to generate. Add your text prompt in the CLIP text encode (prompt) box (use the upper prompt field), then press Run. You’ll be generating AI images fully locally—powered by your AMD RDNA 4 GPU on Windows, without WSL.

Example prompt to try:
beautiful landscape scenery glass bottle with a galaxy inside cute fennec fox snow HDR sunset

This ROCm 7.1.1 update dramatically improves the “plug in and create” experience for Radeon users who want fast, local AI image generation on Windows. If you’ve been waiting for a smoother way to run ComfyUI on AMD hardware, this is the most practical path yet.