Open-source project
hacksider/Deep-Live-Cam avatar
hacksider/Deep-Live-Cam

Deep-Live-Cam 2.8: Real-Time Face Swap with a Single Image, but Mind the License and the Limits

GitHub describes it as real time face swap and one-click video deepfake with only a single image. The repository metadata lists Python as its primary language. The metadata lists the AGPL-3.0 license. This article stays within the project description and details documented in the GitHub repository README.

98,150 stars14,326 forksPythonAGPL-3.0

At a glance

What is it?
Deep-Live-Cam swaps faces in live video using one image, with a GUI and GPU acceleration. It is powerful for creators, but AGPL-3.0 and built-in content checks shape who can use it.
Who is it for?
Adopt Deep-Live-Cam if you are an artist, content creator, or developer who needs real-time face swapping for legitimate projects, you have a CUDA-capable GPU or Apple Silicon, and you can comply with AGPL-3.0 by releasing derivative source code. Do not use it if you need a commercial closed-source product, if you handle sensitive media that the built-in checks might block, or if you cannot manage the manual Python environment.
Can I use it commercially?
Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
Is it still maintained?
Yes. The repository last received commits 7 days ago.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • hacksider/Deep-Live-Cam98,150 ★ · +2,232
Star history
Repository2026-08-132026-11-15
hacksider/Deep-Live-Cam95.9k98.2k

2,232 Stars in the last 94 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

What Deep-Live-Cam Actually Solves

Deep-Live-Cam addresses a specific workflow: swapping a face in a live video stream using a single static image. The README frames it as a tool for the AI-generated media industry, with uses like animating custom characters, creating content, and even clothing design. It is not a general video editor; it is a real-time pipeline that takes a face input, a camera feed, and outputs a swapped video stream. The target user is someone who wants to do this live, not offline, and who is willing to deal with a Python environment or pay for a pre-built package.

The Three-Click Workflow and Its Real-Time Core

The README's TLDR reduces the operation to three steps: select a face, select a camera, and press live. That simplicity hides the underlying real-time requirement. The tool uses ONNX models, specifically inswapper_128_fp16.onnx for the swap and gfpgan-1024.onnx for face restoration. The '128' in the model name suggests a 128x128 input resolution, which is a common compromise for speed. The 'fp16' indicates half-precision floating point, which cuts memory and accelerates inference on compatible GPUs. The real-time claim depends on that model size and precision. The macOS setup installs python-tk, which the README calls required for the GUI. This is not a headless library; it is a desktop application.

Installation: Manual Setup Is Not for Beginners

The README is explicit: installation requires technical skills. You need Python 3.14 recommended, with 3.11 to 3.14 supported, plus pip, git, and ffmpeg. The Windows setup uses a PowerShell command for ffmpeg: iex (irm ffmpeg.tc.ht). Visual Studio 2022 runtimes are required on Windows. After cloning with git clone --depth 1, you must download two model files from Hugging Face and place them in a 'models' folder. Then you create a virtual environment and run pip install -r requirements.txt. For Apple Silicon, you need brew install [email protected] and [email protected]. There is also a documented recovery path for a broken environment: remove venv, recreate it, and reinstall, with a specific fix for gfpgan and basicsr that involves installing BasicSR from a git URL and reinstalling GFPGAN from TencentARC's repo. This is a fragile setup; the README itself pushes the pre-built 'Ultimate' version as the zero-manual-installation alternative.

GPU Acceleration: CUDA and Apple Silicon Paths

For NVIDIA GPUs, the README specifies CUDA Toolkit 12.8.0 and cuDNN v8.9.7 for CUDA 12.x. The cuDNN bin directory must be in PATH. Then you install PyTorch with a cu128 index URL, uninstall any existing onnxruntime, and install onnxruntime-gpu==1.26.0. This is version-pinned, which means future updates to onnxruntime could break compatibility. For CPU-only machines, the README says it is slower but works with python run.py. For Apple Silicon, the setup uses Python 3.14 and [email protected].

Built-In Content Restrictions: A Double-Edged Sword

The README states that a built-in check prevents the program from processing inappropriate media, including nudity, graphic content, and sensitive material like war footage. This is a deliberate safety measure, but it is also a functional constraint. If your legitimate work involves medical imagery, historical war footage, or artistic nudity, the tool may refuse to process it. This is a genuine limitation for researchers or artists working with sensitive but legal content. The project also warns that it may add watermarks or shut down if legally required, so the feature set is not stable.

Licensing and the 'Ultimate' Paid Version

The repository is licensed under AGPL-3.0. That is a strong copyleft license: if you modify the code and provide it over a network, you must offer the source to users. For a face-swap tool that might be used in a SaaS product, this is a major consideration. The README also promotes a 'Pre-built Deep-Live-Cam 2.8 Ultimate' sold through deeplivecam.net, with 30+ exclusive features, performance optimizations, and priority support. It claims to be the only official website and warns against downloading from elsewhere. This creates a dual model: free AGPL code and a paid pre-built distribution. The paid version may include proprietary additions, but the core repository remains AGPL. If you use the free code, you must comply with AGPL; if you buy the Ultimate, you still need to check what license applies to those extras.

Alternatives and the Trade-Off in Approach

A direct alternative is the original DeepFaceLab, which also performs face swaps but is not real-time and typically requires multiple images or a long training phase for a specific face. DeepFaceLab focuses on offline video processing with higher quality, while Deep-Live-Cam trades quality for speed and single-image input. Another alternative is Faceswap, which is also Python-based and uses deep learning, but it is more oriented toward batch processing and requires training a model per face pair. The key difference is that Deep-Live-Cam uses pre-trained ONNX models (inswapper) that work with one image immediately, whereas Faceswap and DeepFaceLab often need you to train or fine-tune. That makes Deep-Live-Cam faster to start but less customizable. If you need high fidelity for a specific person, the training-based tools may be better; if you need instant live swapping, Deep-Live-Cam is the approach.

Maintenance and Upgrade Costs

The repository shows active development, with a 2.8 release in October 2026, 2.7.5-Ultimate in September 2026 and 2.7-ultimate in August 2026. This indicates a fast release cycle, which means you will need to track updates to fix bugs and get new features. However, the manual installation is brittle. The README includes a recovery procedure for a broken venv, which suggests that dependency conflicts are common. The model files are downloaded separately, so you must keep them in sync with the code. If the project changes models, you will need to re-download. The AGPL license also imposes a maintenance cost on distribution: every time you share a modified version, you must provide source. For a solo developer, this is manageable; for a company, it requires legal review.

Editorial conclusion

Adopt Deep-Live-Cam if you are an artist, content creator, or developer who needs real-time face swapping for legitimate projects, you have a CUDA-capable GPU or Apple Silicon, and you can comply with AGPL-3.0 by releasing derivative source code. Do not use it if you need a commercial closed-source product, if you handle sensitive media that the built-in checks might block, or if you cannot manage the manual Python environment. Before adopting, verify the model licenses, test the content filter with your intended inputs, and confirm that your distribution channel can handle AGPL-3.0 obligations. The project's own README warns that quickstart builds come from a single official site; treat other sources as untrusted.

Frequently asked questions

how to install deep live cam on windows 10

The manual path needs Python 3.11 to 3.14 with 3.14 recommended, pip, git, ffmpeg and the Visual Studio 2022 Runtimes, then a shallow clone, the two ONNX model files placed in a models folder, and a venv with `pip install -r requirements.txt`. The file warns the manual install requires technical skills and is not for beginners.

how to use deep live cam on mac

Apple Silicon from M1 through M5 has its own setup: `brew install [email protected]`, `brew install [email protected]` for the GUI, then a 3.14 virtual environment and `pip install -r requirements.txt`. The dependency file gives non-arm64 Macs a different onnxruntime pin than arm64 ones.

Is Deep Live Cam safe?

The project states a built-in check prevents processing nudity, graphic content, and sensitive material like war footage, and that users must obtain consent before using a real person's face and label output as a deepfake. It also says the project is not responsible for end-user actions, and may shut down the project or add watermarks if legally required.

how to use deep live cam on pc

The three-step flow is: select a face, select which camera to use, press live. Without a GPU you run it with `python run.py`, and the file notes that initial execution will download models of about 300MB.

how to download and use deep live cam

Pre-built builds are offered for Windows, Mac Silicon, CPU, NVIDIA and AMD through deeplivecam.net, which the file names as the single official website alongside this repository. The manual route is the shallow git clone, the two model files from Hugging Face, and a venv install from requirements.txt.

Official sources

  1. Official documentation
  2. Official README
  3. Project repository
  4. Release notes
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/hacksider-deep-live-cam.svg)](https://hysenlabs.com/projects/hacksider-deep-live-cam)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/hacksider-deep-live-cam.svg)](https://hysenlabs.com/projects/hacksider-deep-live-cam)

Was this useful?