Metrixel 2026.2.0 — Linux Preview, SDF export, and a command-line interface

Metrixel 2026.2.0 is our biggest release since launch. We're adding Linux as a Preview platform, shipping a brand-new signed-distance-field (SDF) export for machine-learning pipelines, and bringing the full Metrixel processing pipeline to the command line so you can run it on training servers and CI runners.

If you're building image-to-3D models, shape representations, or simulation-ready assets, this release is for you.

What's new

Signed-distance-field (SDF) export (Professional plan)

Generate dense SDF volumes directly from any imported mesh. Each export produces:

• a raw float32 volume that drops straight into PyTorch, JAX, or your framework of choice,
• a JSON manifest with the volume's bounds, resolution, and provenance,
• a consistent voxel grid you can stack into a training dataset without per-asset normalisation glue.

SDF export runs on your GPU through Vulkan compute, with an automatic CPU fallback when needed. A new flag — --sdf_backend=auto|cpu|vulkan — lets you pin the backend explicitly for reproducible runs.

If you've been writing your own meshes-to-SDF preprocessing for a 3D-ML project, this replaces that whole step.

Linux support (Preview)

Metrixel now runs natively on Linux as a Preview release for early-adopter 3D-ML teams. We ship a single relocatable tarball (Metrixel-2026.2.0-linux-x64.tar.gz.zip) that covers Ubuntu 24.04 and Red Hat Enterprise Linux 9 (x86_64): unzip the wrapper, extract the tarball anywhere, run ./install.sh for application-menu integration, and you're done. On Linux, Metrixel uses your GPU's native Vulkan driver, so SDF generation and rendering run at full bandwidth on discrete NVIDIA GPUs — noticeably faster than on Apple Silicon for large volumes.

The tarball bundles EntVista Hub (the IAM/SSO daemon every EVS desktop product talks to). One download installs both — no need to grab a separate EntVista Hub tarball unless you're staying on Metrixel 2026.2.0 while patching to a newer Hub.

Linux 2026.2.0 is an early-access build for people building Linux 3D-ML pipelines alongside us; we'll graduate it to a general-availability release once we've gathered feedback. macOS and Windows remain the recommended platforms for production deployments.

We ship the tarball on Linux — extract it anywhere and you're done. It's the distribution format for Linux, with no separate package or installer to chase.

Command-line interface (preview)

Metrixel CLI runs the entire processing pipeline without opening the GUI — ideal for Amazon SageMaker training jobs, CI runners, and remote Linux hosts you only reach over SSH. Single-file mode and batch (folder) mode are both supported, and every export option available in the GUI is available on the command line.

The CLI signs in through your existing EntVista Hub session, so there's no separate login flow. Start EntVista Hub once on the same host, and metrixel --mode cli picks up your credentials automatically.

This is the first Metrixel CLI release. Flag names and output formats may evolve in 2026.3.x as we incorporate feedback from early adopters — production use is supported, but pin to a specific Metrixel version in your CI pipelines while we iterate.

Silent / unattended install (all platforms)

To match the headless command-line interface, every Metrixel artifact — the macOS and Windows installers, and the Linux tarball — supports a fully unattended install. No display server, no clicks. Drop it into a training-job bootstrap script, a CI/CD pipeline, a container image build, or a server provisioning step and it lands in the directory you choose with a status code your automation can check.

Linux (tarball — bundles EntVista Hub):

unzip Metrixel-2026.2.0-linux-x64.tar.gz.zip
tar xzf Metrixel-2026.2.0-linux-x64.tar.gz -C /opt
/opt/Metrixel-2026.2.0-linux-x64/install.sh        # wires both metrixel + entvistahub
echo "exit: $?"
# Headless / SageMaker / CI:
#   /opt/Metrixel-2026.2.0-linux-x64/EntVistaHub/bin/EntVistaHub --mode cli &
#   QT_QPA_PLATFORM=offscreen /opt/Metrixel-2026.2.0-linux-x64/bin/Metrixel --mode cli ...

macOS:

./installer.app/Contents/MacOS/installer \
  --root /Applications/Metrixel \
  --accept-licenses --accept-messages --default-answer --confirm-command \
  install

Windows (PowerShell):

.\installer.exe `
  --root "C:\Program Files\Metrixel" `
  --accept-licenses --accept-messages --default-answer --confirm-command `
  install

The Linux tarball uninstalls just as easily: ./install.sh --uninstall from inside the extracted directory removes the desktop entries, icons, and ~/.local/bin/{metrixel,entvistahub} symlinks; deleting the extracted directory removes the rest. On macOS and Windows, a maintenance tool ships alongside the application and accepts update, remove, and purge with the same flags as install.

Animation-aware SDF generation

When you batch-export an animation, each frame's SDF now reflects the actual skinned pose for that frame, instead of always producing the bind pose. This lets you build training datasets that capture motion, not just a static reference.

Improvements

• Faster, smoother SDF on Apple Silicon. SDF generation on Mac uses 3D tile dispatching to stay responsive on large meshes, so you can keep using your machine while volumes generate in the background.
• Cleaner PyTorch handoff. Mesh tensors (.pt) now carry a self-describing header, so downstream loaders read shape and dtype directly instead of guessing.
• More reliable Windows builds. Improvements to our Windows build tooling so logs and installer manifests render correctly on every system locale.

Known limitations

• Linux 2026.2.0 is a Preview release. It ships as a relocatable tarball; we're collecting feedback from early-adopter Linux teams before promoting it to general availability.
• SDF export is a Professional-plan feature. Free and Standard accounts will see the option, but the export will be skipped at runtime with a subscription message.
• Imports beyond 200 MB are skipped to keep processing predictable. We're raising this limit in a follow-up release as we finish testing memory profiles on very large assets.

Pricing

The SDF + manifest export is included with the Professional plan alongside the existing batch mode, TIFF export, and USD animation export. Same Professional price as before — no tier changes for this release.

Platforms

OS	Architecture	Distribution
Ubuntu 24.04 / RHEL 9 (Preview)	x86_64	Metrixel-2026.2.0-linux-x64.tar.gz.zip (tarball — bundles EntVista Hub)
macOS 14 or later	Apple Silicon	.dmg (Apple-signed)
Windows 11	x86_64	.exe (signed)

Download the latest version from the product page.