Deep Live Cam Studio 2.2.3

Deep Live Cam Studio 2.2.3 Windows Release

This patch release fixes processed video Preview autoplay and playback. Preview
now starts without manual timeline scrubbing, keeps a stable window size, and
shows processed frames in order at the hardware's achievable processing rate.

Downloads

• NVIDIA/CUDA installer: DeepLiveCamStudio-2.2.3-x64-setup.exe
• Installer SHA-256: 70CFFB0DFF4FE4E267D78C29F7FDF7426439E16D6F882A8BFDAA67D47891C204
• AMD/Intel DirectML portable ZIP: DeepLiveCamStudio-2.2.3-DirectML-x64-portable.zip
• DirectML portable SHA-256: D869292D088F627A94A63411502017354641CB35DFEB24528687AA30A1E13CD6
• Corresponding source archive: DeepLiveCamStudio-2.2.3-source-5ba5a656ceba.zip
• Source archive SHA-256: A9B85F6CBB5BB4A5A0366D6C387C46DDF456BD24A6EE7A5A992A63E81377C99F
• Source ref: 5ba5a656cebad95e9cb9a95e3156d408bb3e80f7

Use the CUDA installer for NVIDIA GPUs. Use the DirectML ZIP for AMD or Intel
DirectX 12 GPUs: extract it into a new folder and run
DeepLiveCamStudio.exe. Do not copy one runtime over another.

To update an NVIDIA installation, close Deep Live Cam Studio and run the new
setup EXE. It migrates earlier version-named installations into one stable
per-user directory while preserving downloaded models, settings, and logs.
DirectML remains portable: extract each update into a new folder and delete the
old extracted folder after verifying the new copy.

What Changed

Changed

• Processed target-video Preview starts automatically and includes Play, Pause,
  and responsive timeline seeking.
• Preview reuses one decoder and requests the next frame only after the current
  processed frame is displayed. Slower hardware therefore shows every frame in
  order at its achievable processing rate instead of skipping ahead.

Fixed

• Prevented the Preview window from growing on each displayed frame.
• Removed per-frame decoder reopen/seek overhead and unnecessary temporal-state
  resets during sequential playback.
• Hardened Preview shutdown, worker-error retry, idle-timer restart, missing
  frame-count metadata, and file/live-preview mutual exclusion.

The core Preview fix was physically tested on Windows with a Radeon RX 6900 XT:
autoplay, stable window size, smoother sequential playback without skipped
frames, Pause/Play, seeking, and responsive close all passed. The tester also
noted low achieved FPS; that is recorded as a separate processing-throughput
observation, not interpreted as a 30 FPS cap. This release does not promise
real-time source-rate Preview when face processing is expensive.

The CUDA/DirectML dependency sets, installer behavior, model policy, and OBS
Live Output implementation are unchanged from 2.2.2. Final release assets are
still checked separately for their expected GPU providers before publication.

Source and license

Deep-Live-Cam is licensed under AGPL-3.0. Complete corresponding source for the
exact binary release is attached and identified by the source ref above.

The source archive is produced with:

powershell -ExecutionPolicy Bypass -File build\windows\package_source.ps1 -AppVersion 2.2.3 -GitRef 5ba5a656cebad95e9cb9a95e3156d408bb3e80f7

This release preserves attribution to the original project:

• Original project: https://github.com/hacksider/Deep-Live-Cam
• License: AGPL-3.0

Models are not bundled

The installer intentionally does not include model/checkpoint files. The
DirectML portable ZIP also excludes them because model licenses and
redistribution rights are separate from the application license.

After launching, select Set Up Models, or run:

DeepLiveCamStudioCLI.exe --download-models

Downloaded models are stored under:

%LOCALAPPDATA%\DeepLiveCamStudio\models

Requirements and known limitations

• ffmpeg and ffprobe must be installed separately for video workflows.
• OBS Studio must be installed separately for OBS Virtual Camera workflows.
• The public files are unsigned and may trigger Microsoft Defender
  SmartScreen.
• Processed Preview FPS depends on the selected processors, models, source
  content, and hardware; it may be lower than the source video's frame rate.
• Face-swap and enhancer model families retain their documented
  non-commercial or research-use restrictions; no model weights are
  redistributed here.

Release evidence

This release candidate is not publish-approved by release notes alone;
publish approval is recorded in RELEASE_VERIFICATION.md,
MANUAL_RELEASE_GATES.md, and the signed gate documents included with the
upload.

Completed local evidence is included in the uploaded release documents:

• Full automated test suite.
• Packaged CUDA and DirectML runtime/provider checks.
• Clean-install, legacy-upgrade, registry, shortcut, and uninstall checks.
• Model download/checksum and forbidden-model scans.
• CUDA and DirectML file-render checks.
• Radeon RX 6900 XT DirectML Preview autoplay and playback checks.
• Existing Radeon RX 9060 XT DirectML and NVIDIA RTX 4070 CUDA Live Output
  checks; the Live Output implementation is unchanged in this patch.

Remaining publish blockers:

• None when RELEASE_VERIFICATION.md says Ready to publish without remaining manual gates: YES.

Completed manual signoffs include:

• Clean Windows install without admin rights.
• OBS Virtual Camera and DirectML Live Output workflows.
• Authorized legal review for dependency, model-license, and redistribution obligations.

Deep Live Cam Studio 2.2.2

Deep Live Cam Studio 2.2.2 Windows Release

This maintenance release makes NVIDIA updates behave like a conventional app
upgrade, clarifies the DirectML portable update path, and ships the reviewed
Windows dependency patches merged since 2.2.1.

Downloads

• NVIDIA/CUDA installer: DeepLiveCamStudio-2.2.2-x64-setup.exe
• Installer SHA-256: A86045B96D8BD0CAA6A8D7E8C5EECB70ED095D59D0175590127790D5F953B8A8
• AMD/Intel DirectML portable ZIP: DeepLiveCamStudio-2.2.2-DirectML-x64-portable.zip
• DirectML portable SHA-256: 79C2719CCD7D1D8DB7A943230385A061643739C97CA0FEC162B21A058E663528
• Corresponding source archive: DeepLiveCamStudio-2.2.2-source-20e82f4b8379.zip
• Source archive SHA-256: 5C0813206B5747306CE35003DE403FCE1A2C7D1D4FA947A9BC23CC6910D47CEC
• Source ref: 20e82f4b83790f73f24b7d8b753ca53015ce775e

Use the CUDA installer for NVIDIA GPUs. Use the DirectML ZIP for AMD or Intel
DirectX 12 GPUs: extract it into a new folder and run
DeepLiveCamStudio.exe. Do not copy one runtime over another.

To update an NVIDIA installation, close Deep Live Cam Studio and run the new
setup EXE. It migrates earlier version-named installations into one stable
per-user directory while preserving downloaded models, settings, and logs.
DirectML remains portable: extract each update into a new folder and delete the
old extracted folder after verifying the new copy.

What Changed

Changed

• Standardized NVIDIA installs on
  %LOCALAPPDATA%\Programs\DeepLiveCamStudio instead of a version-named
  directory.
• Updated pip, wheel, PyInstaller, pip-tools, and Ruff in the deterministic
  Windows build and test toolchain.
• Documented the separate NVIDIA installer and DirectML portable update paths.

Fixed

• The NVIDIA installer now removes the registered legacy installation and
  recognizable orphaned version folders before completing the stable upgrade.
• Installer tests now use an isolated app identity and validate a complete
  2.2.1-layout to 2.2.2 migration without disturbing real installations.
• Replaced the unavailable ONNX Runtime GPU 1.24.3 wheel with 1.24.4 and
  refreshed locks and licence evidence.

The unchanged DirectML application path was previously validated on Windows 11
with a Radeon RX 9060 XT: the DirectML badge was active and Preview,
short-video Render, and OBS Live Output all passed with no blocking regression.
The packaged release runtimes are separately checked for their expected GPU
providers; CUDA rendering and Live Output were also verified locally on an
NVIDIA RTX 4070.

Source and license

Deep-Live-Cam is licensed under AGPL-3.0. Complete corresponding source for the
exact binary release is attached and identified by the source ref above.

The source archive is produced with:

powershell -ExecutionPolicy Bypass -File build\windows\package_source.ps1 -AppVersion 2.2.2 -GitRef 20e82f4b83790f73f24b7d8b753ca53015ce775e

This release preserves attribution to the original project:

• Original project: https://github.com/hacksider/Deep-Live-Cam
• License: AGPL-3.0

Models are not bundled

The installer intentionally does not include model/checkpoint files. The
DirectML portable ZIP also excludes them because model licenses and
redistribution rights are separate from the application license.

After launching, select Set Up Models, or run:

DeepLiveCamStudioCLI.exe --download-models

Downloaded models are stored under:

%LOCALAPPDATA%\DeepLiveCamStudio\models

Requirements and known limitations

• ffmpeg and ffprobe must be installed separately for video workflows.
• OBS Studio must be installed separately for OBS Virtual Camera workflows.
• The public files are unsigned and may trigger Microsoft Defender
  SmartScreen.
• Face-swap and enhancer model families retain their documented
  non-commercial or research-use restrictions; no model weights are
  redistributed here.

Release evidence

This release candidate is not publish-approved by release notes alone;
publish approval is recorded in RELEASE_VERIFICATION.md,
MANUAL_RELEASE_GATES.md, and the signed gate documents included with the
upload.

Completed local evidence is included in the uploaded release documents:

• Full automated test suite.
• Packaged CUDA and DirectML runtime/provider checks.
• Clean-install, legacy-upgrade, registry, shortcut, and uninstall checks.
• Model download/checksum and forbidden-model scans.
• CUDA and DirectML file-render checks.
• Radeon RX 9060 XT DirectML and NVIDIA RTX 4070 CUDA Live Output checks.

Remaining publish blockers:

• None when RELEASE_VERIFICATION.md says Ready to publish without remaining manual gates: YES.

Completed manual signoffs include:

• Clean Windows install without admin rights.
• OBS Virtual Camera and DirectML Live Output workflows.
• Authorized legal review for dependency, model-license, and redistribution obligations.

Deep Live Cam Studio 2.2.1

Deep Live Cam Studio 2.2.1 Windows Release

This maintenance release modernizes the Windows dependency set, hardens the
release process, and safely refactors the UI/frame pipeline while preserving
the CUDA and DirectML workflows introduced in 2.2.0.

Downloads

• NVIDIA/CUDA installer: DeepLiveCamStudio-2.2.1-x64-setup.exe
• Installer SHA-256: D643924BFA22BF2B1BB309BE0DD7436B8A7DB8854003C9437950FBB2133752AF
• AMD/Intel DirectML portable ZIP: DeepLiveCamStudio-2.2.1-DirectML-x64-portable.zip
• DirectML portable SHA-256: 5DD825C8801B3F1C3E43C243D13967E5600C135C3076F7A3E423C648B347D5FD
• Corresponding source archive: DeepLiveCamStudio-2.2.1-source-1eebeb4cc0ec.zip
• Source archive SHA-256: BF5138299E0F1F879A9FA525BE9364C9A51C5150A0D15D475F4DB480B2D38A4F
• Source ref: 1eebeb4cc0ec9d1e978182c19778d6f74538ba29

Use the CUDA installer for NVIDIA GPUs. Use the DirectML ZIP for AMD or Intel
DirectX 12 GPUs: extract it into a new folder and run
DeepLiveCamStudio.exe. Do not copy one runtime over another.

What Changed

Changed

• Refreshed the reviewed Windows dependency set and regenerated reproducible
  CUDA and DirectML locks plus licence snapshots.
• Refactored high-complexity UI and frame-pipeline orchestration behind
  characterization tests while preserving user-visible behavior.
• Modernized the repository README, documentation, issue forms, support and
  security policies, dependency policy, and public project presentation.

Fixed

• Removed stale release triggers, obsolete duplicate tooling, unused imports,
  and retired scaffolding without changing supported workflows.
• Kept strict provider checks so CPU fallback cannot be mistaken for a
  successful CUDA or DirectML result.

An independent physical tester validated the release candidate on Windows 11
with a Radeon RX 9060 XT: the DirectML badge was active and Preview,
short-video Render, and OBS Live Output all passed with no blocking regression.
The packaged release runtimes are separately checked for their expected GPU
providers; CUDA rendering and Live Output were also verified locally on an
NVIDIA RTX 4070.

Source and license

Deep-Live-Cam is licensed under AGPL-3.0. Complete corresponding source for the
exact binary release is attached and identified by the source ref above.

The source archive is produced with:

powershell -ExecutionPolicy Bypass -File build\windows\package_source.ps1 -AppVersion 2.2.1 -GitRef 1eebeb4cc0ec9d1e978182c19778d6f74538ba29

This release preserves attribution to the original project:

• Original project: https://github.com/hacksider/Deep-Live-Cam
• License: AGPL-3.0

Models are not bundled

The installer intentionally does not include model/checkpoint files. The
DirectML portable ZIP also excludes them because model licenses and
redistribution rights are separate from the application license.

After launching, select Set Up Models, or run:

DeepLiveCamStudioCLI.exe --download-models

Downloaded models are stored under:

%LOCALAPPDATA%\DeepLiveCamStudio\models

Requirements and known limitations

• ffmpeg and ffprobe must be installed separately for video workflows.
• OBS Studio must be installed separately for OBS Virtual Camera workflows.
• The public files are unsigned and may trigger Microsoft Defender
  SmartScreen.
• Face-swap and enhancer model families retain their documented
  non-commercial or research-use restrictions; no model weights are
  redistributed here.

Release evidence

This release candidate is not publish-approved by release notes alone;
publish approval is recorded in RELEASE_VERIFICATION.md,
MANUAL_RELEASE_GATES.md, and the signed gate documents included with the
upload.

Completed local evidence is included in the uploaded release documents:

• Full automated test suite.
• Packaged CUDA and DirectML runtime/provider checks.
• Installer smoke install/uninstall checks.
• Model download/checksum and forbidden-model scans.
• CUDA and DirectML file-render checks.
• Radeon RX 9060 XT DirectML and NVIDIA RTX 4070 CUDA Live Output checks.

Remaining publish blockers:

• None when RELEASE_VERIFICATION.md says Ready to publish without remaining manual gates: YES.

Completed manual signoffs include:

• Clean Windows install without admin rights.
• OBS Virtual Camera and DirectML Live Output workflows.
• Authorized legal review for dependency, model-license, and redistribution obligations.

Deep Live Cam Studio 2.2.0

Deep Live Cam Studio 2.2.0 Windows Release

This release adds a verified DirectML build for AMD and Intel GPUs and fixes a
file Preview/Render deadlock that could also affect CUDA systems.

Downloads

• NVIDIA/CUDA installer: DeepLiveCamStudio-2.2.0-x64-setup.exe
• Installer SHA-256: 55B83C48BDCDF818E3F051AD2896A1495AED1DEA2E676BD3F8E071A7DC45916E
• AMD/Intel DirectML portable ZIP: DeepLiveCamStudio-2.2.0-DirectML-x64-portable.zip
• DirectML portable SHA-256: 177AD450FB2BBBE34B5B9F0BACFB77D7541BFD16201524B6A4324A02E57A32FB
• Corresponding source archive: DeepLiveCamStudio-2.2.0-source-0e2617e0c944.zip
• Source archive SHA-256: A4BB8F774FEA453AE9077C8F7E3D004B5297DEBF9CE9A9FF0C0CAF56417696D5
• Source ref: 0e2617e0c944f911aa26c144878292e0272270c9

Use the CUDA installer for NVIDIA GPUs. Use the DirectML ZIP for AMD or Intel
DirectX 12 GPUs: extract it into a new folder and run
DeepLiveCamStudio.exe. Do not copy one runtime over another.

What Changed

Added

• DirectML GPU acceleration for AMD and Intel GPUs, including strict provider
  verification and multi-GPU adapter selection.
• A versioned DirectML portable ZIP with a SHA-256 sidecar and release-time
  archive validation.

Fixed

• Preview and Start Render can no longer re-enter while the first file
  operation is loading models, preventing the observed Not Responding
  deadlock.
• DirectML face analysis uses the CPU compatibility path while swap and
  enhancement inference remain on the GPU, avoiding Radeon multi-session
  hangs.
• GitHub packaging now preserves dot-prefixed runtime directories, including
  the scikit-learn .libs DLL required by the application.
• Clean NVIDIA installer builds now include the pinned CUDA 12/cuDNN 9 runtime
  DLL set required for an actual ONNX Runtime CUDA session.
• Provider diagnostics fail clearly when requested GPU acceleration is not
  active instead of silently treating CPU fallback as success.

The final DirectML artifact was validated by the issue reporter on a Radeon
6900 XT for file rendering and Live Output. CUDA rendering and Live Output were
also verified locally on an NVIDIA RTX 4070.

Source and license

Deep-Live-Cam is licensed under AGPL-3.0. Complete corresponding source for the
exact binary release is attached and identified by the source ref above.

The source archive is produced with:

powershell -ExecutionPolicy Bypass -File build\windows\package_source.ps1 -AppVersion 2.2.0 -GitRef 0e2617e0c944f911aa26c144878292e0272270c9

This release preserves attribution to the original project:

• Original project: https://github.com/hacksider/Deep-Live-Cam
• License: AGPL-3.0

Models are not bundled

The installer intentionally does not include model/checkpoint files. The
DirectML portable ZIP also excludes them because model licenses and
redistribution rights are separate from the application license.

After launching, select Set Up Models, or run:

DeepLiveCamStudioCLI.exe --download-models

Downloaded models are stored under:

%LOCALAPPDATA%\DeepLiveCamStudio\models

Requirements and known limitations

• ffmpeg and ffprobe must be installed separately for video workflows.
• OBS Studio must be installed separately for OBS Virtual Camera workflows.
• The public files are unsigned and may trigger Microsoft Defender
  SmartScreen.
• Face-swap and enhancer model families retain their documented
  non-commercial or research-use restrictions; no model weights are
  redistributed here.

Release evidence

This release candidate is not publish-approved by release notes alone;
publish approval is recorded in RELEASE_VERIFICATION.md,
MANUAL_RELEASE_GATES.md, and the signed gate documents included with the
upload.

Completed local evidence is included in the uploaded release documents:

• Full automated test suite.
• Packaged CUDA and DirectML runtime/provider checks.
• Installer smoke install/uninstall checks.
• Model download/checksum and forbidden-model scans.
• CUDA and DirectML file-render checks.
• Radeon 6900 XT and NVIDIA RTX 4070 Live Output checks.

Remaining publish blockers:

• None when RELEASE_VERIFICATION.md says Ready to publish without remaining manual gates: YES.

Completed manual signoffs include:

• Clean Windows install without admin rights.
• OBS Virtual Camera and DirectML Live Output workflows.
• Authorized legal review for dependency, model-license, and redistribution obligations.

Deep Live Cam Studio 2.1.9

Hotfix Notice - May 20, 2026

The Windows installer asset was replaced after the initial v2.1.9 upload to fix a laptop startup issue where the process could close before the UI or desktop launch log appeared.

Current installer SHA-256:

AF804BE93D95A254930E1CFDDCFBA5EA955DFB04A07FCDCB6BC94EF4A7B0B334  DeepLiveCamStudio-2.1.9-x64-setup.exe

After installing, DeepLiveCamStudio.exe should have SHA-256:

D38DE56BE5162D164F6048FFB08D8B36F1DAD8578E1B918907C758120DDE3DE8

If you downloaded v2.1.9 before this notice, download the installer again.

---

# Deep Live Cam Studio 2.1.9 Windows Release

This release packages a modified build of Deep-Live-Cam for Windows x64.

Downloads

• Windows installer: DeepLiveCamStudio-2.1.9-x64-setup.exe
• Installer SHA-256: EFF4D46727AD8157A0527AAC30797B6BDB030E83002483C98CBA617029CC1F0A
• Corresponding source archive: DeepLiveCamStudio-2.1.9-source-c676426c8eb8.zip
• Source archive SHA-256: CFEA4A8CF07B25641750DFB99600442BB98FC855D923A8E6F67546CC1AE7065E
• Source ref: c676426c8eb8f1bd61d17439eece02d2426065bd

Source And License

Deep-Live-Cam is licensed under AGPL-3.0. This binary release is distributed with corresponding source availability requirements.

Attach the source archive listed above to the GitHub Release, or link the exact release tag/commit in the publishing repository. The public release must make the complete corresponding source available for the exact binary installer.

The source archive must be produced from the release Git ref with:

powershell -ExecutionPolicy Bypass -File build\windows\package_source.ps1 -AppVersion 2.1.9 -GitRef c676426c8eb8f1bd61d17439eece02d2426065bd

This release preserves attribution to the original project:

• Original project: https://github.com/hacksider/Deep-Live-Cam
• License: AGPL-3.0

Models Are Not Bundled

The installer intentionally does not include model/checkpoint files (.onnx, .pth, .safetensors) because model licenses and redistribution rights are separate from the application license.

After installing, users can review model sources, license notes, and SHA-256 checksums before downloading:

DeepLiveCamStudioCLI.exe --download-models

Downloaded models are stored under:

%LOCALAPPDATA%\DeepLiveCamStudio\models

Notable Packaging Changes

• Windows x64 per-user installer using Inno Setup.
• PyInstaller onedir app bundle; Python does not need to be installed by users.
• Start menu shortcut and optional desktop shortcut.
• Clean uninstall of app files; user model data is preserved during silent uninstall and interactive uninstall asks before removing models.
• WebP and AVIF source-image upload support.
• Default-on extended subject blend mask for smoother hairline, neck, shoulder, and upper-chest compositing.
• Layered extended subject matte with soft falloff to reduce hard mask edges around hair and shoulders.
• New Studio controls for extended subject masking and subject-mask preview overlay.
• Fixed alpha-channel preview frames before face detection and swap-model inference.
• Packaged runtime preflight and installer smoke tests are included under build/windows.

Known Requirements

• ffmpeg/ffprobe must be installed separately for video workflows unless a future release explicitly bundles a reviewed ffmpeg build.
• NVIDIA GPU acceleration requires compatible NVIDIA drivers and runtime libraries. CPU fallback should be tested before publishing.
• OBS Virtual Camera workflows require OBS Studio installed separately.

Compliance Notes

Review these files before publishing:

• COMPLIANCE.md
• THIRD_PARTY_NOTICES.md
• LICENSES/PYTHON_DEPENDENCIES.md
• MODEL_DOWNLOAD_VERIFICATION.md
• PROCESSING_VERIFICATION.md
• RELEASE_CHECKLIST.md
• RELEASE_REPORT.md

Known legal risks that must not be hidden from release notes:

• The inswapper model family has historical non-commercial/research-use restrictions from InsightFace even where mirrors are labeled GPL-3.0.
• GPEN upscaler model mirrors are marked non-commercial, academic, and educational use only.
• pyvirtualcam metadata reports GPLv2; review binary distribution compatibility.
• Dependency metadata can be incomplete and should be legally reviewed before public distribution.

Pre-Publish Checks

This release candidate is not publish-approved by release notes alone; publish
approval is recorded in RELEASE_VERIFICATION.md, MANUAL_RELEASE_GATES.md,
and the signed gate documents included with the upload.

Completed local evidence is included in the uploaded release documents:

• Packaged runtime preflight.
• Installer smoke test.
• Model download/checksum verification.
• CPU fallback processing.
• CUDA processing on the local NVIDIA test machine.

Remaining publish blockers:

• None when RELEASE_VERIFICATION.md says Ready to publish without remaining manual gates: YES.

Completed manual signoffs include:

• Clean Windows VM install without admin rights.
• OBS Virtual Camera workflow.
• Authorized legal review for dependency, model-license, and redistribution obligations.

Deep Live Cam Studio 2.1.8

Deep Live Cam Studio 2.1.8 Windows Release

This release packages a modified build of Deep-Live-Cam for Windows x64.

Downloads

• Windows installer: DeepLiveCamStudio-2.1.8-x64-setup.exe
• Installer SHA-256: A5B3075CACDAEA68694FB26907BB10F3C3A92F3FB7278F429073B5F9A8DAC2F1
• Corresponding source archive: DeepLiveCamStudio-2.1.8-source-a3944bcd9b98.zip
• Source archive SHA-256: 8E71D8F01BD1EB75664BC515AED2C455C5DCDF75588A96371C2628DDFBA94CC5
• Source ref: a3944bcd9b98017f6ca43e130495f2652fb0d2d3

Source And License

Deep-Live-Cam is licensed under AGPL-3.0. This binary release is distributed with corresponding source availability requirements.

Attach the source archive listed above to the GitHub Release, or link the exact release tag/commit in the publishing repository. The public release must make the complete corresponding source available for the exact binary installer.

The source archive must be produced from the release Git ref with:

powershell -ExecutionPolicy Bypass -File build\windows\package_source.ps1 -AppVersion 2.1.8 -GitRef a3944bcd9b98017f6ca43e130495f2652fb0d2d3

This release preserves attribution to the original project:

• Original project: https://github.com/hacksider/Deep-Live-Cam
• License: AGPL-3.0

Models Are Not Bundled

The installer intentionally does not include model/checkpoint files (.onnx, .pth, .safetensors) because model licenses and redistribution rights are separate from the application license.

After installing, users can review model sources, license notes, and SHA-256 checksums before downloading:

DeepLiveCamStudioCLI.exe --download-models

Downloaded models are stored under:

%LOCALAPPDATA%\DeepLiveCamStudio\models

Notable Packaging Changes

• Windows x64 per-user installer using Inno Setup.
• PyInstaller onedir app bundle; Python does not need to be installed by users.
• Start menu shortcut and optional desktop shortcut.
• Clean uninstall of app files; user model data is preserved during silent uninstall and interactive uninstall asks before removing models.
• WebP and AVIF source-image upload support.
• Cleaner Studio UI media controls with the unreliable Random face action removed.
• Responsive restored-window layout for the header, media slots, and render controls.
• Fixed Start Live so it sends frames to OBS Virtual Camera instead of only opening the preview window.
• Fixed the Studio Exit button so it stops child windows, cleans up, and closes the app reliably.
• Dependency security refresh for ONNX, Pillow, and Protobuf; see SECURITY_REVIEW.md.
• Packaged runtime preflight and installer smoke tests are included under build/windows.

Known Requirements

• ffmpeg/ffprobe must be installed separately for video workflows unless a future release explicitly bundles a reviewed ffmpeg build.
• NVIDIA GPU acceleration requires compatible NVIDIA drivers and runtime libraries. CPU fallback should be tested before publishing.
• OBS Virtual Camera workflows require OBS Studio installed separately.

Compliance Notes

Review these files before publishing:

• COMPLIANCE.md
• THIRD_PARTY_NOTICES.md
• LICENSES/PYTHON_DEPENDENCIES.md
• MODEL_DOWNLOAD_VERIFICATION.md
• PROCESSING_VERIFICATION.md
• RELEASE_CHECKLIST.md
• RELEASE_REPORT.md

Known legal risks that must not be hidden from release notes:

• The inswapper model family has historical non-commercial/research-use restrictions from InsightFace even where mirrors are labeled GPL-3.0.
• GPEN upscaler model mirrors are marked non-commercial, academic, and educational use only.
• pyvirtualcam metadata reports GPLv2; review binary distribution compatibility.
• Dependency metadata can be incomplete and should be legally reviewed before public distribution.

Pre-Publish Checks

This release candidate is not publish-approved by release notes alone; publish
approval is recorded in RELEASE_VERIFICATION.md, MANUAL_RELEASE_GATES.md,
and the signed gate documents included with the upload.

Completed local evidence is included in the uploaded release documents:

• Packaged runtime preflight.
• Installer smoke test.
• Model download/checksum verification.
• CPU fallback processing.
• CUDA processing on the local NVIDIA test machine.

Remaining publish blockers:

• None when RELEASE_VERIFICATION.md says Ready to publish without remaining manual gates: YES.

Completed manual signoffs include:

• Clean Windows VM install without admin rights.
• OBS Virtual Camera workflow.
• Authorized legal review for dependency, model-license, and redistribution obligations.

Deep Live Cam Studio 2.1.7

Deep Live Cam Studio 2.1.7 Windows Release

This release packages a modified build of Deep-Live-Cam for Windows x64.

Downloads

• Windows installer: DeepLiveCamStudio-2.1.7-x64-setup.exe
• Installer SHA-256: 07DAD2B8A589BC79712B18A5A544469FDC483150A81EA55654531FECE5339FA5
• Corresponding source archive: DeepLiveCamStudio-2.1.7-source-96d5577a9709.zip
• Source archive SHA-256: 209F8BFBC98758823DC1EE0F2356235330682F6A7C9556E5D7FC0180F4ADD63E
• Source ref: 96d5577a97091581c7dd76352a79bbed057cd6fe

Source And License

Deep-Live-Cam is licensed under AGPL-3.0. This binary release is distributed with corresponding source availability requirements.

Attach the source archive listed above to the GitHub Release, or link the exact release tag/commit in the publishing repository. The public release must make the complete corresponding source available for the exact binary installer.

The source archive must be produced from the release Git ref with:

powershell -ExecutionPolicy Bypass -File build\windows\package_source.ps1 -AppVersion 2.1.7 -GitRef 96d5577a97091581c7dd76352a79bbed057cd6fe

This release preserves attribution to the original project:

• Original project: https://github.com/hacksider/Deep-Live-Cam
• License: AGPL-3.0

Models Are Not Bundled

The installer intentionally does not include model/checkpoint files (.onnx, .pth, .safetensors) because model licenses and redistribution rights are separate from the application license.

After installing, users can review model sources, license notes, and SHA-256 checksums before downloading:

DeepLiveCamStudioCLI.exe --download-models

Downloaded models are stored under:

%LOCALAPPDATA%\DeepLiveCamStudio\models

Notable Packaging Changes

• Windows x64 per-user installer using Inno Setup.
• PyInstaller onedir app bundle; Python does not need to be installed by users.
• Start menu shortcut and optional desktop shortcut.
• Clean uninstall of app files; user model data is preserved during silent uninstall and interactive uninstall asks before removing models.
• WebP and AVIF source-image upload support.
• Cleaner Studio UI media controls with the unreliable Random face action removed.
• Responsive restored-window layout for the header, media slots, and render controls.
• Dependency security refresh for ONNX, Pillow, and Protobuf; see SECURITY_REVIEW.md.
• Packaged runtime preflight and installer smoke tests are included under build/windows.

Known Requirements

• ffmpeg/ffprobe must be installed separately for video workflows unless a future release explicitly bundles a reviewed ffmpeg build.
• NVIDIA GPU acceleration requires compatible NVIDIA drivers and runtime libraries. CPU fallback should be tested before publishing.
• OBS Virtual Camera workflows require OBS Studio installed separately.

Compliance Notes

Review these files before publishing:

• COMPLIANCE.md
• THIRD_PARTY_NOTICES.md
• LICENSES/PYTHON_DEPENDENCIES.md
• MODEL_DOWNLOAD_VERIFICATION.md
• PROCESSING_VERIFICATION.md
• RELEASE_CHECKLIST.md
• RELEASE_REPORT.md

Known legal risks that must not be hidden from release notes:

• The inswapper model family has historical non-commercial/research-use restrictions from InsightFace even where mirrors are labeled GPL-3.0.
• GPEN upscaler model mirrors are marked non-commercial, academic, and educational use only.
• pyvirtualcam metadata reports GPLv2; review binary distribution compatibility.
• Dependency metadata can be incomplete and should be legally reviewed before public distribution.

Pre-Publish Checks

This release candidate is not publish-approved by release notes alone; publish
approval is recorded in RELEASE_VERIFICATION.md, MANUAL_RELEASE_GATES.md,
and the signed gate documents included with the upload.

Completed local evidence is included in the uploaded release documents:

• Packaged runtime preflight.
• Installer smoke test.
• Model download/checksum verification.
• CPU fallback processing.
• CUDA processing on the local NVIDIA test machine.

Remaining publish blockers:

• None when RELEASE_VERIFICATION.md says Ready to publish without remaining manual gates: YES.

Completed manual signoffs include:

• Clean Windows VM install without admin rights.
• OBS Virtual Camera workflow.
• Authorized legal review for dependency, model-license, and redistribution obligations.

Deep Live Cam Studio 2.1.6

Deep Live Cam Studio 2.1.6 Windows Release

This release packages a modified build of Deep-Live-Cam for Windows x64.

Downloads

• Windows installer: DeepLiveCamStudio-2.1.6-x64-setup.exe
• Installer SHA-256: listed in the uploaded RELEASE_ASSETS.md and DeepLiveCamStudio-2.1.6-x64-setup.exe.sha256
• Corresponding source archive: DeepLiveCamStudio-2.1.6-source-c1d5ccc79432.zip
• Source archive SHA-256: 92D77E068306522BE237A65EB788C195564E9BFD2B995E7790DE8AE28E829B66
• Source ref: c1d5ccc79432fad3d9757ba2906b15de41f390c8

Source And License

Deep-Live-Cam is licensed under AGPL-3.0. This binary release is distributed with corresponding source availability requirements.

Attach the source archive listed above to the GitHub Release, or link the exact release tag/commit in the publishing repository. The public release must make the complete corresponding source available for the exact binary installer.

The source archive must be produced from the release Git ref with:

powershell -ExecutionPolicy Bypass -File build\windows\package_source.ps1 -AppVersion 2.1.6 -GitRef HEAD

This release preserves attribution to the original project:

• Original project: https://github.com/hacksider/Deep-Live-Cam
• License: AGPL-3.0

Models Are Not Bundled

The installer intentionally does not include model/checkpoint files (.onnx, .pth, .safetensors) because model licenses and redistribution rights are separate from the application license.

After installing, users can review model sources, license notes, and SHA-256 checksums before downloading:

DeepLiveCamStudioCLI.exe --download-models

Downloaded models are stored under:

%LOCALAPPDATA%\DeepLiveCamStudio\models

Notable Packaging Changes

• Windows x64 per-user installer using Inno Setup.
• PyInstaller onedir app bundle; Python does not need to be installed by users.
• Start menu shortcut and optional desktop shortcut.
• Clean uninstall of app files; user model data is preserved during silent uninstall and interactive uninstall asks before removing models.
• WebP and AVIF source-image upload support.
• Packaged runtime preflight and installer smoke tests are included under build/windows.

Known Requirements

• ffmpeg/ffprobe must be installed separately for video workflows unless a future release explicitly bundles a reviewed ffmpeg build.
• NVIDIA GPU acceleration requires compatible NVIDIA drivers. The Windows installer bundles the CUDA 12/cuDNN 9 runtime DLLs needed by onnxruntime-gpu; CPU fallback should be tested before publishing.
• OBS Virtual Camera workflows require OBS Studio installed separately.

Compliance Notes

Review these files before publishing:

• COMPLIANCE.md
• THIRD_PARTY_NOTICES.md
• LICENSES/PYTHON_DEPENDENCIES.md
• MODEL_DOWNLOAD_VERIFICATION.md
• PROCESSING_VERIFICATION.md
• RELEASE_CHECKLIST.md
• RELEASE_REPORT.md

Known legal risks that must not be hidden from release notes:

• The inswapper model family has historical non-commercial/research-use restrictions from InsightFace even where mirrors are labeled GPL-3.0.
• GPEN upscaler model mirrors are marked non-commercial, academic, and educational use only.
• pyvirtualcam metadata reports GPLv2; review binary distribution compatibility.
• Dependency metadata can be incomplete and should be legally reviewed before public distribution.

Pre-Publish Checks

This release candidate is not publish-approved by release notes alone; publish
approval is recorded in RELEASE_VERIFICATION.md, MANUAL_RELEASE_GATES.md,
and the signed gate documents included with the upload.

Completed local evidence is included in the uploaded release documents:

• Packaged runtime preflight.
• Installer smoke test.
• Model download/checksum verification.
• CPU fallback processing.
• CUDA processing on the local NVIDIA test machine.

Remaining publish blockers:

• None when RELEASE_VERIFICATION.md says Ready to publish without remaining manual gates: YES.

Completed manual signoffs include:

• Clean Windows VM install without admin rights.
• OBS Virtual Camera workflow.
• Authorized legal review for dependency, model-license, and redistribution obligations.

Deep Live Cam Studio 2.1.5 Windows x64

Deep Live Cam Studio 2.1.5 Windows Release

This release packages a modified build of Deep-Live-Cam for Windows x64.

Downloads

• Windows installer: DeepLiveCamStudio-2.1.5-x64-setup.exe
• Installer SHA-256: C904A7930F7643313669EA5A20A7027772D8FB126BD33756073683A48B7BBE7F
• Corresponding source archive: DeepLiveCamStudio-2.1.5-source-5c238d709b27.zip
• Source archive SHA-256: E62B27F97845907AD25A429067E0C604A3E0DD2DB05C6A0C0DDCD4402949ABB3
• Source ref: 5c238d709b2721cc72b16b5a6c21ada0657c5078

Source And License

Deep-Live-Cam is licensed under AGPL-3.0. This binary release is distributed with corresponding source availability requirements.

Attach the source archive listed above to the GitHub Release, or link the exact release tag/commit in the publishing repository. The public release must make the complete corresponding source available for the exact binary installer.

The source archive must be produced from the release Git ref with:

powershell -ExecutionPolicy Bypass -File build\windows\package_source.ps1 -AppVersion 2.1.5 -GitRef 5c238d709b2721cc72b16b5a6c21ada0657c5078

This release preserves attribution to the original project:

• Original project: https://github.com/hacksider/Deep-Live-Cam
• License: AGPL-3.0

Models Are Not Bundled

The installer intentionally does not include model/checkpoint files (.onnx, .pth, .safetensors) because model licenses and redistribution rights are separate from the application license.

After installing, users can review model sources, license notes, and SHA-256 checksums before downloading:

DeepLiveCamStudioCLI.exe --download-models

Downloaded models are stored under:

%LOCALAPPDATA%\DeepLiveCamStudio\models

Notable Packaging Changes

• Windows x64 per-user installer using Inno Setup.
• PyInstaller onedir app bundle; Python does not need to be installed by users.
• Start menu shortcut and optional desktop shortcut.
• Clean uninstall of app files; user model data is preserved during silent uninstall and interactive uninstall asks before removing models.
• WebP and AVIF source-image upload support.
• Packaged runtime preflight and installer smoke tests are included under build/windows.

Known Requirements

• ffmpeg/ffprobe must be installed separately for video workflows unless a future release explicitly bundles a reviewed ffmpeg build.
• NVIDIA GPU acceleration requires compatible NVIDIA drivers and runtime libraries. CPU fallback should be tested before publishing.
• OBS Virtual Camera workflows require OBS Studio installed separately.

Compliance Notes

Review these files before publishing:

• COMPLIANCE.md
• THIRD_PARTY_NOTICES.md
• LICENSES/PYTHON_DEPENDENCIES.md
• MODEL_DOWNLOAD_VERIFICATION.md
• PROCESSING_VERIFICATION.md
• RELEASE_CHECKLIST.md
• RELEASE_REPORT.md

Known legal risks that must not be hidden from release notes:

• The inswapper model family has historical non-commercial/research-use restrictions from InsightFace even where mirrors are labeled GPL-3.0.
• GPEN upscaler model mirrors are marked non-commercial, academic, and educational use only.
• pyvirtualcam metadata reports GPLv2; review binary distribution compatibility.
• Dependency metadata can be incomplete and should be legally reviewed before public distribution.

Pre-Publish Checks

This release candidate is not publish-approved by release notes alone; publish
approval is recorded in RELEASE_VERIFICATION.md, MANUAL_RELEASE_GATES.md,
and the signed gate documents included with the upload.

Completed local evidence is included in the uploaded release documents:

• Packaged runtime preflight.
• Installer smoke test.
• Model download/checksum verification.
• CPU fallback processing.
• CUDA processing on the local NVIDIA test machine.

Remaining publish blockers:

• None when RELEASE_VERIFICATION.md says Ready to publish without remaining manual gates: YES.

Completed manual signoffs include:

• Clean Windows VM install without admin rights.
• OBS Virtual Camera workflow.
• Authorized legal review for dependency, model-license, and redistribution obligations.