Autolume Metacreation Lab

generative-ai model-crafting no-code small-data stylegan

Use this command to install Autolume with WinGet:

winget install --id=MetacreationLab.Autolume -e

No-code generative AI for artists to craft real-time visual performances

Autolume

Autolume is a cross-platform tool for real-time visual performances using StyleGAN.

For detailed documentation and demos look at the Metacreation Lab Website.

Autolume Screen

Installation

➡️ Download for Windows, macOS and Linux

Winget (Windows)

Install the app

winget install MetacreationLab.Autolume

Update the app

winget upgrade MetacreationLab.Autolume

Brew (macOS)

Add and trust the brew tap (one time)

brew tap metacreation-lab/tap
brew trust metacreation-lab/tap

Install the app

brew install --cask autolume

Update the app

brew upgrade --cask autolume

Examples

For examples of artworks created with Autolume see: https://www.metacreation.net/artworks

Development instructions

uv is required for setting up the python environment (installation link)

Dependencies

Windows

CUDA 12.8 (download link)
- Minimum components: CUDA Development + CUDA Runtime
Microsoft C++ Build Tools (download link)
- Minimum components: Desktop development with C++

Linux (Ubuntu 24+)

CUDA (download link)
- Minimum components: CUDA Development + CUDA Runtime
PortAudio runtime (for audio input): sudo apt install libportaudio2

macOS

Command Line Tools for Xcode: xcode-select --install

On macOS and Windows, PortAudio ships inside the sounddevice wheel, so no extra install is needed for audio input.

Initial setup

Create the Python environment:

uv sync

Running the program

Start the program

uv run main.py

Benchmarking

Measure a generator's batch-1 forward latency (auto-selects CUDA, MPS, or CPU)

uv run bench.py path/to/network.pkl

Create a release

Run the cross-platform release script on the OS you want to build for (PyInstaller cannot cross-compile — a Windows build must run on Windows, etc.):

uv run release.py

The script requires a .env at the repo root with the crash report endpoint values (it fails fast without one). Copy .env.example and fill it in — see tools/crash_endpoint/README.md. Forks without their own endpoint can build with --disable-crash-reporting instead.

Windows / Linux: output is the dist/Autolume/ folder.
macOS: output is dist/Autolume.app, code-signed automatically — see macOS signing and notarization.

Pass --package to additionally produce a distributable: an .AppImage on Linux, a .dmg on macOS, or an Inno Setup installer plus a portable .tar.xz archive on Windows (requires Inno Setup on the build machine, installable with winget install --id JRSoftware.InnoSetup.7 -e -s winget -i). Use --package-only to package an existing dist/ output without rebuilding.

macOS signing and notarization

Gatekeeper only opens a downloaded app without friction when it is both signed with a Developer ID and notarized by Apple (an automated malware scan of the uploaded file — this applies to all distribution outside the App Store, GitHub releases included). The release script handles both, after a one-time setup.

One-time setup

Signing certificate: install a Developer ID Application certificate and its private key in your login keychain. Verify it is visible with:
```
security find-identity -v -p codesigning
```
Notarization credentials: ask a team admin for an App Store Connect API key with the Developer role (App Store Connect → Users and Access → Integrations → Team Keys). You need three values from that page: the .p8 key file, the Key ID, and the Issuer ID. Store them under the autolume-notary keychain profile:
```
xcrun notarytool store-credentials autolume-notary \
  --key AuthKey_XXXXXX.p8 --key-id  --issuer-id 
```

Release flow

uv run release.py --package --notarize

The build signs Autolume.app (hardened runtime + entitlements.plist) with the Developer ID certificate found in the keychain. If there is none, the script warns and falls back to an ad-hoc signature; the app then still runs locally, but downloads are blocked by Gatekeeper.
--package wraps the app into a .dmg and signs it.
--notarize submits the .dmg to Apple (usually takes a few minutes) and staples the resulting ticket, so users can open the app with a plain double-click. It can also be run on its own against an already-built .dmg — notarizing after the fact also unblocks copies that were published or downloaded earlier, since Gatekeeper checks Apple's servers online.

Building documentation

uv run zensical build

The documentation will be output to site.

To serve the documentation locally:

uv run zensical serve

The documentation will be served at http://127.0.0.1:8000/.

Credits

Philippe Pasquier: Director of Metacreation Lab for Creative AI, Principal Investigator
Arshia Sobhan: Project management
Lionel Ringenbach: Software development
Michael Tjokrowardojo: Software development
Jonas Kraasch: Project first-phase development
Arthur Deleu: Project development
Mahshid Jabari: DevOps
Peter Chen: Contributions to the super-resolution module
Manuel Bolduc: User testing
Ahmed Abuzuraiq: Research associate
Anthony Hong: macOS port research and testing

Acknowledgment

StyleGAN3: https://github.com/NVlabs/stylegan3

GAN compression: https://github.com/lychenyoko/content-aware-gan-compression

GANSpace: https://github.com/harskish/ganspace

License

Some of the dependencies fall under the Nvidia Source Code License.

Project status

This code base is still open to changes and bugs might still appear. Please create issues and let us know so we can polish it for the final release!

As a dedicated research team behind Autolume, we emphasize that our system is a tool for exploration and innovation without direct support.

Autolume

Autolume is a cross-platform tool for real-time visual performances using StyleGAN.

For detailed documentation and demos look at the Metacreation Lab Website.

Autolume Screen

Installation

➡️ Download for Windows, macOS and Linux

Winget (Windows)

Install the app

winget install MetacreationLab.Autolume

Update the app

winget upgrade MetacreationLab.Autolume

Brew (macOS)

Add and trust the brew tap (one time)

brew tap metacreation-lab/tap
brew trust metacreation-lab/tap

Install the app

brew install --cask autolume

Update the app

brew upgrade --cask autolume

Examples

For examples of artworks created with Autolume see: https://www.metacreation.net/artworks

Development instructions

uv is required for setting up the python environment (installation link)

Dependencies

Windows

CUDA 12.8 (download link)
- Minimum components: CUDA Development + CUDA Runtime
Microsoft C++ Build Tools (download link)
- Minimum components: Desktop development with C++

Linux (Ubuntu 24+)

CUDA (download link)
- Minimum components: CUDA Development + CUDA Runtime
PortAudio runtime (for audio input): sudo apt install libportaudio2

macOS

Command Line Tools for Xcode: xcode-select --install

On macOS and Windows, PortAudio ships inside the sounddevice wheel, so no extra install is needed for audio input.

Initial setup

Create the Python environment:

uv sync

Running the program

Start the program

uv run main.py

Benchmarking

Measure a generator's batch-1 forward latency (auto-selects CUDA, MPS, or CPU)

uv run bench.py path/to/network.pkl

Create a release

Run the cross-platform release script on the OS you want to build for (PyInstaller cannot cross-compile — a Windows build must run on Windows, etc.):

uv run release.py

Windows / Linux: output is the dist/Autolume/ folder.
macOS: output is dist/Autolume.app, code-signed automatically — see macOS signing and notarization.

macOS signing and notarization

One-time setup

Signing certificate: install a Developer ID Application certificate and its private key in your login keychain. Verify it is visible with:
```
security find-identity -v -p codesigning
```
Notarization credentials: ask a team admin for an App Store Connect API key with the Developer role (App Store Connect → Users and Access → Integrations → Team Keys). You need three values from that page: the .p8 key file, the Key ID, and the Issuer ID. Store them under the autolume-notary keychain profile:
```
xcrun notarytool store-credentials autolume-notary \
  --key AuthKey_XXXXXX.p8 --key-id  --issuer-id 
```

Release flow

uv run release.py --package --notarize

The build signs Autolume.app (hardened runtime + entitlements.plist) with the Developer ID certificate found in the keychain. If there is none, the script warns and falls back to an ad-hoc signature; the app then still runs locally, but downloads are blocked by Gatekeeper.
--package wraps the app into a .dmg and signs it.
--notarize submits the .dmg to Apple (usually takes a few minutes) and staples the resulting ticket, so users can open the app with a plain double-click. It can also be run on its own against an already-built .dmg — notarizing after the fact also unblocks copies that were published or downloaded earlier, since Gatekeeper checks Apple's servers online.

Building documentation

uv run zensical build

The documentation will be output to site.

To serve the documentation locally:

uv run zensical serve

The documentation will be served at http://127.0.0.1:8000/.

Credits

Philippe Pasquier: Director of Metacreation Lab for Creative AI, Principal Investigator
Arshia Sobhan: Project management
Lionel Ringenbach: Software development
Michael Tjokrowardojo: Software development
Jonas Kraasch: Project first-phase development
Arthur Deleu: Project development
Mahshid Jabari: DevOps
Peter Chen: Contributions to the super-resolution module
Manuel Bolduc: User testing
Ahmed Abuzuraiq: Research associate
Anthony Hong: macOS port research and testing

Acknowledgment

StyleGAN3: https://github.com/NVlabs/stylegan3

GAN compression: https://github.com/lychenyoko/content-aware-gan-compression

GANSpace: https://github.com/harskish/ganspace

License

Some of the dependencies fall under the Nvidia Source Code License.

Project status

This code base is still open to changes and bugs might still appear. Please create issues and let us know so we can polish it for the final release!

As a dedicated research team behind Autolume, we emphasize that our system is a tool for exploration and innovation without direct support.