How to Set Up WAN Video Generation on Windows 11 with LoRA [Wan2gp]
Wan2GP install on Windows 11 [tips and potential fixes]
I recently needed to setup Wan2GP for a client project and if you’ve got an NVIDIA GPU with decent VRAM and want to get into AI video generation, WAN (Wan2.1/2.2) models are currently some of the best open-source options available. Here’s my guide to getting it running on Windows, including the gotchas I ran into along the way.
What You’ll Need
- Windows 10/11
- NVIDIA GPU (I’m running a 5090 mobile with 24GB VRAM, but 12GB+ should work with optimization settings)
- About 20-30GB free disk space for models
- Some patience for first-run model downloads
Your Two Main Options
Option 1: Wan2GP (Recommended for Most People)
Wan2GP is an open-source tool created by developer DeepBeepMeep that provides a clean web UI for running WAN models. Key advantages:
- Automatic model downloading on first use
- Built-in LoRA support
- Supports Wan 2.1/2.2, Hunyuan Video, LTX Video, and Flux
- One-click installer available
Option 2: ComfyUI + Kijai’s WanVideoWrapper
More flexible with a node-based workflow, better if you want complex pipelines or already use ComfyUI. Kijai’s implementation stays on the cutting edge with optimization features.
For this guide, I’m focusing on Wan2GP since it’s the path of least resistance.
Prerequisites (Don’t Skip These!)
Before running any installer, you need these components. Trust me, the installer will fail with vague errors if any are missing.
1. Git for Windows
Download from: https://git-scm.com/downloads/win
Run the installer with default settings. After installation, open a Command Prompt and verify it works:
git --version
You should see something like git version 2.52.0.windows.1.
2. NVIDIA Drivers
Get the latest from the NVIDIA App or: https://www.nvidia.com/drivers
Open a Command Prompt and verify:
nvidia-smi
You should see a table with your GPU info and driver version.
3. Visual Studio Build Tools with C++
Some acceleration libraries (SageAttention, Triton) need to compile CUDA kernels during installation. This requires the C++ build tools.
Download from: https://visualstudio.microsoft.com/downloads/
Scroll down to “Tools for Visual Studio” and download Build Tools for Visual Studio.
Note about versions: The Wan2GP readme mentions Visual Studio 2022, but as of late 2025, the download page shows Visual Studio 2026. That’s fine — VS 2017, 2019, 2022, and 2026 share the same redistributable files and are binary compatible. Just download whatever current version is available.
During installation:
- The Visual Studio Installer will open
- Select the “Desktop development with C++” workload (check the checkbox)
- On the right panel, make sure MSVC build tools and Windows SDK are checked
- Click Install
This is a several GB download, but you’re only getting the build tools — not the full Visual Studio IDE.
4. CUDA Toolkit (This One Tripped Me Up)
Important: Having NVIDIA drivers that “support” CUDA is not the same as having the CUDA Toolkit installed. When you run nvidia-smi, you’ll see a “CUDA Version” listed — that’s just telling you what version your driver supports. You still need to install the actual toolkit for compiling CUDA code.
Download from: https://developer.nvidia.com/cuda-toolkit
During installation:
- Express install works fine for most people
- If doing Custom, you mainly need: CUDA → Development → Compiler, Libraries, and Runtime
- You can skip GeForce Experience, Nsight tools, etc.
You may see a warning that Visual Studio 2022 or 2019 wasn’t detected (because you installed VS 2026). This is fine — just check the box that says “I understand, and wish to continue the installation regardless” and proceed. The Nsight debugging/profiling tools are optional and not needed for Wan2GP.
After installation, close any open Command Prompt windows and open a fresh one. Then verify CUDA is working:
nvcc --version
You should see output like:
nvcc: NVIDIA (R) Cuda compiler driver
Cuda compilation tools, release 13.1, V13.1.xxx
If You Get “‘nvcc’ is not recognized”
The installer didn’t add CUDA to your system PATH. Here’s how to fix it manually:
- Press
Win + R, typesysdm.cpl, press Enter - Go to the Advanced tab
- Click Environment Variables
- Under “System variables”, find Path and click Edit
- Click New and add the path to your CUDA bin folder. It will be something like:
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\bin(Adjust the version number to match what you installed) - Click OK on all dialogs to save
- Open a brand new Command Prompt (this is important — existing windows won’t see the change)
- Test again with
nvcc --version
Installing Wan2GP
One-Click Installer (Easiest)
The official Wan2GP repo recommends the Redtash1 One-Click Installer. It’s a repackaged portable version with pre-compiled dependencies and has the main developer’s endorsement.
Steps:
- Go to the GitHub releases page and download the latest
.exefile - Double-click the
.exe— it’s a self-extracting archive (7-Zip SFX) - Choose where to extract. Use a simple path without spaces — something like
C:\Wan2GPworks great. Paths with spaces can break batch scripts. - After extraction, navigate to the folder and run the appropriate Install batch file for your GPU:
Install_RTX_50XX.batfor RTX 5000 series (Blackwell)Install_RTX_40XX.batfor RTX 4000 series- And so on for other GPU generations
The installer will download dependencies and set everything up. This takes a while.
When installation completes, you’ll have a desktop shortcut to launch Wan2GP — no need to mess with command prompts or batch files for everyday use.
Manual Install (More Control)
If you prefer managing your own Python environment, open a Command Prompt and run:
git clone https://github.com/deepbeepmeep/Wan2GP.git
cd Wan2GP
conda create -n wan2gp python=3.10.9
conda activate wan2gp
pip install torch==2.6.0+cu126 torchvision==0.21.0+cu126 torchaudio==2.6.0+cu126 --index-url https://download.pytorch.org/whl/cu126
pip install -r requirements.txt
pip install triton-windows
pip install sageattention==1.0.6
Then launch with:
python wgp.py
First Run
Whether you used the one-click installer or manual method, the web interface will be available at:
http://127.0.0.1:7860
Open that URL in your browser.
Heads up: First run will be slow. Wan2GP auto-downloads models when you first select them, and they’re several GB each. Grab a coffee.
Using LoRAs
Wan2GP has built-in LoRA support. Check the folder structure in your Wan2GP installation directory — there are organized subfolders for different model types. The web UI also has guidance on where to place LoRA files for each model you’re using.
Tips
- With 24GB VRAM, you can run the 14B models at full quality without aggressive memory optimization
- Check the Guides tab in the web UI for recommended models and settings
- The community Discord is active if you run into model-specific issues
Troubleshooting
“System checks failed” with no clear error
The Wan2GP installer gives a frustratingly vague error message when prerequisites are missing. To figure out what’s actually wrong, open a Command Prompt and run these one at a time:
git --version
nvidia-smi
nvcc --version
echo %CUDA_PATH%
Whichever one fails or returns an error is your culprit.
‘nvcc’ is not recognized after CUDA install
See the PATH fix in the CUDA Toolkit section above. The key thing to remember: always open a fresh Command Prompt after modifying PATH variables. Existing windows won’t see the change.
Visual Studio warning during CUDA install
If the CUDA installer warns that it can’t find Visual Studio 2022/2019, that’s fine. Check the acknowledgment box and continue. The Nsight debugging tools are optional and not required for Wan2GP.
Installer fails immediately with path errors
Move your Wan2GP folder to a simple path without spaces, like C:\Wan2GP. Batch scripts often choke on paths like C:\Users\Name\Downloads\Wan2GP One Click Install\.
Windows Defender flags the installer
The one-click installer is a self-extracting 7-Zip archive, which can trigger false positives. Since it’s officially recommended by the Wan2GP developer, it should be safe — but if you’re cautious, the manual conda installation method gives you full control over every step.
Wrapping Up
Once you’re past the prerequisites hurdle, Wan2GP makes video generation surprisingly accessible. The web UI is intuitive, model management is automatic, and with enough VRAM you can generate impressive results without fighting memory constraints.
PS these are just some notes meant to help, if I made a mistake in writing things down please accept my apologies. But hopefully, something here may help you get started yourself!
Happy generating!