How to Install FaceFusion on Windows, RunPod and Massed Compute (1-Click)
By Furkan Gözükara · Updated Nov 21, 2026
The easiest FaceFusion install on Windows, RunPod or Massed Compute is the SECourses 1-click FaceFusion installer: on Windows you install the requirements once, run Windows_Install.bat and start the app with Windows_Start_App.bat; on the cloud you pick the right template or image and follow the instructions file. The current installer sets up FaceFusion 3.5.2 and uses uv, which made installs about 100 times faster on RunPod.
What the installer covers
| Platform | How it installs | Notes |
|---|---|---|
| Windows (local) | Windows_Install.bat, then Windows_Start_App.bat | Install the Windows requirements first |
| RunPod | Runpod_Instructions_READ.txt | Use the template written in that file |
| Massed Compute | Massed_Compute_Instructions_READ.txt | Select the SECourses image from the Creator dropdown |
The installer zip (FaceFusion_v18.zip at the time of the January 2026 update) is available to members on Patreon in the FaceFusion 1-click installers post. It works on GPUs from the RTX 1000 to the RTX 5000 series and on server GPUs, and the app even starts on CPU if you have no decent GPU. In testing on an RTX 3090 Ti and an RTX 5090, the RTX 5090 reached 60 FPS, a record in those tests.
Windows requirements
FaceFusion depends on a correct CUDA and cuDNN setup. Since the January 2026 update, the Windows installer no longer downloads cuDNN and TensorRT for you, so install these first:
- Python 3.10 (the Windows installer uses 3.10.11)
- CUDA 12.9
- cuDNN 9.12
- FFmpeg
- C++ build tools and MSVC
- Git
The Windows AI requirements tutorial shows every step, including how to install and switch CUDA versions, and its written companion, the CUDA and C++ tools fix guide, is kept up to date. If you prefer an automated setup, see the 1-click AI requirements installer guide.
Install FaceFusion on Windows step by step
- Install the requirements above and set the CUDA path exactly as the tutorial shows.
- Download the latest installer zip and extract it into a new folder. A fresh install into a new folder is recommended over updating an old one.
- Double-click
Windows_Install.batand let it finish. - Start the app with
Windows_Start_App.bat. - Check the console output. You should see lines like:
Using CUDA from: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9Using cuDNN from: C:\Program Files\NVIDIA\CUDNN\v9.12\bin\12.9
- To update later, run
Windows_Update.bat.
If those two lines do not appear, go back to the requirements tutorial and install everything exactly as shown before trying again.
Fixing common install errors
The requirements tutorial installs FaceFusion 3 as part of the video and walks through the errors it hit, so it doubles as a troubleshooting guide:
- "CUDA_PATH is set but CUDA wasn't able to be loaded": an ONNX Runtime error that means CUDA and cuDNN are not installed the way the app expects. The tutorial has a dedicated chapter at 24:49.
- DLL or ONNX Runtime errors when FaceFusion starts: shown and solved at 33:31. That 2024 fix copied cuDNN 9.4 DLL files; the current installer expects the versions listed above.
- Git cloning errors: covered at 30:31.
- C++ build or compile errors: see how to fix CUDA and C++ build errors for Python AI apps.
Older guides for FaceFusion NEXT listed CUDA 11.8; the legacy FaceFusion NEXT post is kept only for reference, and the maintained installer is the one above.
Install FaceFusion on RunPod
- Deploy a pod with the template written in
Runpod_Instructions_READ.txt. In the September 2025 update this was the RunPod PyTorch 2.8.0 template,runpod/pytorch:2.8.0-py3.11-cuda12.8.1-cudnn-devel-ubuntu22.04. - Run the commands from the instructions file. With the right template, the install is fully automatic.
The general RunPod workflow for SECourses installers starts at 22:03 in this cloud install tutorial. In May 2025, installs and model downloads on RunPod were reported as extremely slow; the move to uv in January 2026 made the installer about 100 times faster there. The Deep Live Cam post names FaceFusion as its recommended face swap app for RunPod.
Install FaceFusion on Massed Compute
- Register on Massed Compute and use the coupon SECourses.
- Deploy an instance with the SECourses image from the Creator dropdown.
- Follow
Massed_Compute_Instructions_READ.txt.
The Massed Compute part of the same cloud tutorial starts at 12:58. The FaceFusion post recommends Massed Compute as the cloud option, describing it as ultra fast for installation and model downloads. To compare providers, see RunPod vs Massed Compute vs SimplePod.
CUDA or TensorRT?
Use CUDA. TensorRT compilation takes over 10 minutes and turned out slower than CUDA in testing, so it is not recommended, although you can still try it. On RunPod and Massed Compute, TensorRT is not recommended either because of the long compile time, while CUDA runs fast.
FaceFusion vs other face swap apps
| App | Best for | Platforms |
|---|---|---|
| FaceFusion 3.5.2 | All-round deepfake and face swap app with many extra features | Windows, RunPod, Massed Compute |
| VisoMaster Fusion | Face swap studio for images, video, webcam and VR180, with DeepFaceLab DFM model support | Windows (also portable, no requirements), Massed Compute and Linux |
| Deep Live Cam | Live webcam face swap; weaker than FaceFusion and VisoMaster for other uses | Windows |
| Rope Live and Alucard | Advanced face swap with a one-time TensorRT compile | Windows, Massed Compute, Linux |
| Roop Unleashed | Easy face swap with live webcam support | Windows, RunPod, Massed Compute |
All of these installers are available to members on Patreon. For Massed Compute, the Deep Live Cam post recommends either FaceFusion or VisoMaster. VisoMaster Fusion's Windows_Start_Portable.bat installs everything into a portable Python environment, which helps if you cannot get the Windows requirements working. Whatever app you use, only swap the faces of people who have agreed to it.
FAQ
Does FaceFusion need an NVIDIA GPU?
The installer supports GPUs from the RTX 1000 to the RTX 5000 series and server GPUs, and FaceFusion can even start on CPU if you have no decent GPU. For speed, a GPU with CUDA is the intended setup; an RTX 5090 reached 60 FPS.
Why does FaceFusion not find CUDA or cuDNN?
Usually the wrong versions are installed or the path is not set. When the app starts correctly, it prints where it found CUDA 12.9 and cuDNN 9.12. If it does not, follow the requirements tutorial step by step, including its chapter on the "CUDA_PATH is set but CUDA wasn't able to be loaded" error.
Should I update FaceFusion or make a fresh install?
A fresh install into a new folder is recommended, especially after big installer changes. For routine updates on Windows, Windows_Update.bat is included.
Is RunPod or Massed Compute better for FaceFusion?
The FaceFusion post recommends Massed Compute because installs and model downloads were much faster there. RunPod works too with the template from the instructions file, and the uv-based installer made RunPod installs about 100 times faster.