| name | setup-environment |
| description | Use when setting up a fresh clone of this repo to run OVRTX Dev Variant Presenter (also called the "variant studio" in the walkthrough video and earlier releases) — no .venv yet, ovrtx/ovstage/ovstream not installed, ModuleNotFoundError on ovrtx/ovstage/ovstream, or "how do I install / get this running". RTX + uv; validated on Windows. |
Set up the environment to run OVRTX Dev Variant Presenter
The environment is NOT shipped — .venv is gitignored. uv sync rebuilds it from pyproject.toml.
Prerequisites (README §Requirements)
- An NVIDIA RTX GPU + recent driver. Rendering is headless via Vulkan; no display server needed. This app does not run without an RTX GPU. Developed and validated on Windows 11; Linux is supported by the stack (all three packages ship manylinux wheels) but is not routinely validated here.
- Python 3.11, via
uv. uv must already be installed (it is NOT auto-installed here).
- A modern Chromium browser (Chrome / Edge) for the viewer.
- A USD stage to open — none is bundled; the public NVIDIA ConceptCar URL in the README is the quickest one to try.
Install
From the repo root:
uv sync
This creates .venv and installs everything — including ovrtx (renderer), ovstage
(live stage host), and ovstream (WebRTC streamer) — from PyPI; uv.lock pins the
validated versions (ovrtx 0.4.x, ovstage 0.1.x, ovstream 0.4.x, Python 3.11). No
extra index, flag, or auth needs to be configured — but the install does need network
access to pypi.nvidia.com: the ovrtx package on PyPI is only a wheel-stub sdist
whose build backend downloads the real platform wheel from NVIDIA's index, so an environment
restricted to pypi.org alone cannot complete uv sync. (run_server.ps1 /
run_server.sh also run this automatically on first launch if .venv is missing.)
One browser-side file is not in the repo and is not installed by uv sync (the wheels
ship no JavaScript): web/omniverse-webrtc-streaming-library.js, NVIDIA's StreamSDK WebRTC
client, which the viewer cannot stream without. It is NVIDIA's own software, so it is fetched
rather than redistributed: on first launch run_server.ps1 / run_server.sh download it from
the public NVIDIA-Omniverse/ovstream
repository at a pinned commit and verify its SHA256, deleting the file and refusing to start
if the checksum does not match. This needs network access to raw.githubusercontent.com, once —
if the file is already on disk, no download happens.
Verify
uv run python -c "import ovrtx, ovstage, ovstream, warp; print('env OK')"
Next
Start the server with the start-server skill — always via the watchdog (run_server.ps1 on Windows, run_server.sh on Linux/macOS), never bare python or a generic preview/dev-server launcher.
If it fails
uv: command not found → install uv first (see its docs); not handled here.- ovrtx / ovstage / ovstream won't resolve → confirm Python 3.11 (
ovrtx ships no 3.12+ builds) and network access to both pypi.org and pypi.nvidia.com (the ovrtx PyPI package is a wheel-stub sdist that fetches the real wheel from NVIDIA's index at build time — a pypi.org-only mirror or an offline proxy fails here).
- Import fails at the verify step → wrong/no GPU+driver. An RTX GPU is required; the validated configuration is Windows + RTX.
- The launcher aborts on the stream-client fetch (network blocked / offline / SHA256 mismatch) → download
https://raw.githubusercontent.com/NVIDIA-Omniverse/ovstream/af7f1f9006d1037a3cc7b8eca73f39a6469b69c2/examples/webrtc_client/omniverse-webrtc-streaming-library.js on any machine that can reach raw.githubusercontent.com, save it as web/omniverse-webrtc-streaming-library.js in the repo, and relaunch (the launcher skips the fetch when the file exists). The page also says so itself: if that file is missing the viewer renders an explanatory message instead of a dead page.
