Video Generation

Faceswap

Free, open-source deepfake software for training face-swap models on your own GPU

Free and open source (GPL-3.0); optional donations via Patreon/PayPal
Visit Faceswap →
Pricing
Free and open source (GPL-3.0); optional donations via Patreon/PayPal
Best for
Good for developers, researchers and VFX hobbyists with a decent GPU and patience.
Official site
faceswap.dev
Last checked
2026-10-29

Faceswap is great for technically comfortable hobbyists, VFX tinkerers and researchers who want a free, local, trainable face-swap pipeline; skip it if you want instant results, a hosted app, or any vendor support. It is a volunteer-run, GPL-3.0 open-source project built on Python, TensorFlow and Keras with three stages: extract faces from source media, train a model on two face sets, then convert target video or images. It offers both a GUI and a command-line interface, several model architectures, and runs on Windows, macOS and Linux. Note that the official site's latest news post is dated April 2020 and the GitHub Releases section is empty, so how actively it is maintained today could not be verified; the repository shows a 2026 copyright notice and 1,899 commits. The project publishes an ethics statement against non-consensual use, and using any deepfake tool on people without consent can be illegal.

Best for

Good for developers, researchers and VFX hobbyists with a decent GPU and patience. Not for non-technical users wanting one-click swaps, anyone without a capable GPU, teams needing support or an SLA, or anyone intending to make non-consensual content.

Key features

01

Extract

Detects and crops faces from a folder of source photos or video frames into a dedicated extract folder, ready for training.

02

Train

Trains a custom neural network model from two folders of source and destination face images, saving progress into a models folder; existing models can be reused to speed up training.

03

Convert

Applies a trained model to swap faces into a target set of images or video frames, producing the final output.

04

GUI and CLI

Offers both a graphical interface (python faceswap.py gui) and full command-line control with help flags on every script.

05

Multiple Model Architectures

Supports several swappable model types including Villain, DFL-H128, DFaker, Phaze-A, Unbalanced, and OHR, each with different quality/speed tradeoffs.

06

GPU Acceleration

Leverages CUDA-capable GPUs for training speed, with ROCm support extending GPU acceleration to many AMD cards on Linux.

07

ffmpeg Conversion Tool

Includes a bundled effmpeg utility to convert video into photo sequences and back into video for the extract/convert pipeline.

08

Community Support Channels

An active Discord server and dedicated forum provide troubleshooting help, guides, and tutorials outside the GitHub issue tracker.

Pricing breakdown

Open Source (Self-Hosted)

$0
Free under GPL-3.0; optional Patreon/PayPal donations
  • Full source code and all features
  • Runs on your own Windows, macOS or Linux hardware
  • No credits, caps or subscription
  • Your own hardware and electricity are the real cost

Pros and cons

Pros

  • No licence cost: GPL-3.0 code, no subscription or per-generation fee; the cost is your hardware.
  • Everything runs locally, so source footage is not uploaded to a third-party service.
  • Extract/Train/Convert is well documented in the repository and forums, with a Discord and forum community.
  • Several model architectures (Villain, DFL-H128, DFaker, Phaze-A and more, per the project docs) let you trade quality against speed.
  • Cross-platform with AMD ROCm support on Linux, so not NVIDIA-only.

Cons

  • Expect real setup work: Python environment, GPU drivers and a multi-step workflow.
  • Good results typically need a strong GPU and hours to days of training.
  • No vendor support; troubleshooting goes through forums and Discord, and the GitHub tracker is not for general questions.
  • Activity level could not be verified: old site news (April 2020), no GitHub releases listed, so check recent commits and compatibility with your GPU/driver stack before investing time.
  • Deepfake tech carries legal and ethical risk; misuse can be unlawful in many places.

Frequently asked questions

Is Faceswap really free?

Yes. The software is GPL-3.0 licensed with no paid tier. The site accepts optional donations via Patreon and PayPal. Your cost is hardware and electricity.

What hardware do I need?

The repository recommends a modern NVIDIA GPU with CUDA; many AMD GPUs work through ROCm on Linux. Exact minimum specs were not listed on the pages we checked.

Is it still maintained?

Could not be verified. The repository shows 9 open issues and a 2026 copyright notice, but the website's latest news is from April 2020 and no GitHub releases are listed.

Is it an easy online face-swap app?

No. It is a local training tool; results take training time and technical setup. For quick hosted face edits look at browser-based tools instead.

Ready to try Faceswap?

Head to the official site to explore pricing and start a free trial where available.

Visit Faceswap →