Deep Live Cam Studio

A Windows face-swap studio for file rendering and live camera output, with dedicated CUDA and DirectML builds.

NVIDIA: choose the setup .exe  ·  AMD or Intel: choose the DirectML .zip

Download · Quick start · Features · Documentation · Contribute · Security

Important

Select Download for Windows above, then expand Assets if necessary. Do not use the green Code button—its source archives do not install the application.

Caution

Use face-swap software only with consent and for lawful purposes. Do not use it for impersonation, fraud, harassment, non-consensual sexual content, or misleading media.

Download

Open the latest release, expand Assets, and choose the build that matches your hardware:

Hardware	Download	Packaging
NVIDIA GPU	DeepLiveCamStudio-<version>-x64-setup.exe	Per-user Windows installer with CUDA runtime support
AMD or Intel GPU	DeepLiveCamStudio-<version>-DirectML-x64-portable.zip	Extract-and-run DirectML package
Developers	Corresponding-source archive	Exact source and packaging scripts for the release

DirectML also works on supported NVIDIA GPUs, but CUDA is the recommended NVIDIA profile.

Release assets include SHA-256 sidecars and a combined SHA256SUMS.txt. Windows may show Microsoft Defender SmartScreen because the public installer does not yet have paid code-signing reputation. Proceed only when the file came from this repository and its checksum matches the release page.

Updating

• NVIDIA/CUDA installer: Close Deep Live Cam Studio and run the newer setup .exe. It replaces the registered installation, migrates older version-named install folders to one stable location, and keeps downloaded models, settings, and logs.
• AMD/Intel DirectML portable: Close the app and extract the newer ZIP into a new folder. After it starts successfully, delete the previous extracted folder. Models, settings, and logs remain under %LOCALAPPDATA%\DeepLiveCamStudio.

Do not extract the DirectML package over the CUDA installation or mix files from the two packages.

Features

Capability	What it provides
Windows desktop studio	A native PySide6 interface for source selection, preview, rendering, refinement, and live output.
NVIDIA acceleration	A packaged CUDA installer with the reviewed runtime libraries required by ONNX Runtime GPU.
AMD and Intel acceleration	A separate DirectML portable build for DirectX 12-capable Windows GPUs.
OBS and virtual cameras	Process a camera feed and send it to OBS, meeting apps, or other virtual-camera consumers.
Consent-based model setup	Models are excluded from releases and downloaded only after the user reviews their source, licence notes, and checksums.
Reproducible releases	Versioned dependency locks, SHA-256 sidecars, corresponding-source archives, and automated release checks.

Quick Start

1. Download the CUDA installer for NVIDIA or the DirectML ZIP for AMD/Intel.
2. Run the NVIDIA installer, or extract the DirectML ZIP into a new folder.
3. Start Deep Live Cam Studio.
4. Select Set Up Models and review the model sources, licence notes, and checksums.
5. Select a source face and target image, video, or camera.
6. Use Preview, Start Render, or Start Live.

Video processing requires ffmpeg and ffprobe:

winget install Gyan.FFmpeg

Close and reopen the app after installation so the new commands are available.

Requirements

Requirement	Notes
Operating system	64-bit Windows 10 22H2, Windows 11 23H2, or newer
NVIDIA profile	Supported NVIDIA GPU and a current production driver
DirectML profile	AMD, Intel, or NVIDIA DirectX 12-capable GPU with a current vendor driver
Video files	ffmpeg and ffprobe on PATH
Live output	Camera access and an installed virtual-camera consumer such as OBS
Models	Downloaded separately with explicit consent; not bundled in releases

See DirectML testing and OBS Virtual Camera for provider-specific setup and troubleshooting.

Using the Studio

Files

Choose a source face and target image or video. Video previews start automatically; use Play/Pause or the timeline to inspect individual frames. Adjust the quality and refinement controls, then select Start Render. Video output retains audio when ffmpeg is available and Keep audio is enabled.

Live Output

Select a camera, configure the processing options, and choose Start Live. The processed feed can be consumed by OBS or another virtual-camera application. Stop live output before changing providers or closing the app.

Command Line

The packaged app includes a diagnostic and automation CLI:

DeepLiveCamStudioCLI.exe --check-execution-provider --execution-provider cuda
DeepLiveCamStudioCLI.exe --source source.png --target target.png --output output.png --execution-provider cuda

DirectML users can select a particular adapter:

DeepLiveCamStudioCLI.exe --check-execution-provider --execution-provider directml --directml-device-id 0

Provider checks fail when the requested accelerator is unavailable instead of silently treating CPU fallback as success.

Models and Privacy

• Face-swap and enhancement models are not bundled with the installer or portable archive.
• Model downloads require explicit consent and are checked against reviewed SHA-256 values.
• Media stays on the local machine unless the user moves or shares it.
• Treat faces, media files, third-party models, and logs as sensitive or untrusted input.
• Never attach private faces, videos, credentials, or model binaries to a public GitHub issue.

Model sources and redistribution constraints are documented in the model licence audit.

Documentation

Start with the documentation index.

Users

• DirectML setup and provider verification
• OBS and virtual-camera setup
• Support and troubleshooting routes
• Security policy

Developers and Maintainers

• Build from source
• Dependency locks and supported runtime matrix
• Contribution guide
• Release checklist
• Governance and maintenance policy
• Licence and release compliance

Build From Source

Use Python 3.11 on Windows. The CUDA and DirectML profiles are intentionally isolated because their ONNX Runtime packages conflict.

py -3.11 -m venv venv
venv\Scripts\python.exe -m pip install -r requirements.txt -r requirements-dev.txt
venv\Scripts\python.exe run.py --execution-provider cuda

For DirectML, packaged builds, tests, and release commands, follow docs/BUILDING.md.

Security and Release Integrity

• Report vulnerabilities through GitHub private vulnerability reporting, never a public issue.
• Download only from this repository's release page.
• Verify release hashes before installing on a sensitive system.
• Public binary releases include the corresponding source and licence evidence required by AGPL-3.0.

See SECURITY.md for supported versions, response targets, and reporting guidance.

Contributing

Issues and pull requests are welcome when they are reproducible, scoped, and safe. Hardware-facing changes must distinguish automated provider checks from physical CUDA/DirectML/OBS validation.

Read CONTRIBUTING.md and the Code of Conduct before contributing. Use GitHub Discussions for questions and ideas; use issues for confirmed bugs and planned work.

Project Scope and Attribution

This repository is a Windows-focused derivative of hacksider/Deep-Live-Cam. It prioritizes packaged Windows releases, reproducible dependency profiles, hardware-provider validation, and OBS live output. Upstream history and contributors remain part of the project history.

Licence and Compliance

The application is licensed under GNU AGPL-3.0. Model files and third-party components may have separate terms. Anyone distributing a modified binary must provide the complete corresponding source for that binary and preserve applicable notices.

Review COMPLIANCE.md, THIRD_PARTY_NOTICES.md, and the licence evidence index before redistributing the application.