Generated banner for Omniverse Realtime Viewer
Generated logo for Omniverse Realtime Viewer

Omniverse Realtime Viewer

NVIDIAOfficialCreated by NVIDIA Omniverse

  • Skill
  • Tool integration
  • Software development
  • Apache-2.0
  • Medium risk

From the author

Use as the top-level router for Omniverse Realtime Viewer USD apps, XR delivery, SPG GPU processing, stylized rendering, and focused viewer reference documents.
— the description by NVIDIA Omniverse

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

The Omniverse Realtime Viewer skill acts as a central routing and architectural framework for developing USD-based 3D viewer applications. It provides a structured approach to composing rendering, scene state, and delivery mechanisms, ensuring that developers can integrate OVRTX rendering, OVStage scene management, and various UI or streaming adapters effectively. The skill emphasizes a modular architecture where the viewer runtime, UI/input adapters, and delivery systems are decoupled, allowing for consistent behavior across local and streamed environments.

What it does

  • Provides a routing architecture for USD viewer apps, XR delivery, and GPU-accelerated rendering.
  • Defines the separation of concerns between viewer runtime, UI adapters, and delivery systems.
  • Offers guidance on integrating OVRTX for high-fidelity 3D rendering.
  • Establishes conventions for camera interaction, scene state management, and input handling.
  • Supports the implementation of specialized features like stylized rendering and CAE/CFD visualization.
  • Provides protocols for agent-to-UI communication and scene control.

When to use it

  • When starting the development of a new USD-based realtime viewer application.
  • When integrating advanced rendering features like stylized visualization or AOV switching.
  • When building XR or spatial editing capabilities into an existing Omniverse application.
  • When you need to standardize the architecture for local and streamed viewer deployments.
  • When implementing chat-driven scene control or agent-based inspection of a 3D scene.

Advantages

  • Promotes a clean, modular architecture that separates rendering logic from UI and transport.
  • Ensures consistent behavior across different delivery methods, such as local windows or WebRTC streaming.
  • Provides clear paths for extending functionality without disrupting the core viewer runtime.
  • Facilitates the integration of complex simulation and visualization data into the viewer.
  • Standardizes conventions for scene state, selection, and camera control.

Usage examples

“How do I set up a basic USD viewer application using the Omniverse Realtime Viewer architecture?”
The agent will guide you to start with the viewer app reference, resolve dependencies, and implement the core runtime owner to manage the OVStage and OVRTX renderer.
“I need to add stylized rendering to my existing viewer. Where should I start?”
The agent will direct you to the stylized visualization references to implement the required USD/Lua/CUDA nodes within your existing runtime owner.
“How can I enable an AI agent to control the viewer scene?”
The agent will point to the A2UI Scene Copilot reference to implement an operation interface that allows the agent to send commands through the existing runtime owner.
“How do I switch between different render outputs like beauty and depth in my viewer?”
The agent will explain how to use the AOV switching reference to map RenderVars and update the RenderProduct's orderedVars through the runtime owner.

Install

  1. Download the skill as a ZIP:

    https://autogineer.com/agent-skills/nvidia/omniverse-realtime-viewer/download
  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip omniverse-realtime-viewer.zip -d .agents/skills

Documentation by NVIDIA Omniverse

<!-- SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. -->

<!-- SPDX-License-Identifier: Apache-2.0 -->

Omniverse Realtime Viewer

Build responsive USD viewers by composing OVRTX rendering, OVStage scene state, and the delivery/UI libraries appropriate to the app. Use this package for viewer architecture and interaction; use each selected SDK's upstream guidance for installation and exact release-specific APIs.

Start With The App

  1. When delivery is explicit, select its recipe through references/routing.md and follow its current phase. Otherwise start with references/usd-viewer-app/README.md. For an existing app, keep its architecture and read only the references for the requested change. Broad app, conventions and UI catalogs are lookups as needed.

  2. Resolve the selected libraries through references/dependencies/README.md. Preserve the project's manifest unless an upgrade is requested.

  3. Build a working frame and camera interaction in the chosen shell, then add the requested capabilities. Read references/conventions.md for shared input, camera, selection, and protocol conventions.

  4. Use references/viewer-ux-workflow/README.md for broader interface design. Loading, selection feedback, errors, and input responsiveness belong in the first working app, alongside rendering.

  5. Validate the actual delivery path with references/validation.md. A browser screenshot does not validate a native window or a different transport.

Architecture

OwnerResponsibility
Viewer runtimeOwn the live OVStage, publication clock, OVRTX renderer, scene changes, and render stepping.
UI and input adapterDisplay frames and copied app state; send user intent to the runtime owner.
Storage/query/simulation workersPerform their assigned work and return results through the selected integration boundary.
Delivery adapterPresent locally or stream through ovstream; respect frame ownership and backpressure.

Use OVRTX for USD/3D rendering. Browser apps display its WebRTC video; local apps present its output through their native shell. A missing GPU means rendering is unverified, not a reason to substitute a browser 3D renderer.

Keep viewer cameras, settings, effects, and transient edits in app-owned session or runtime state so the source USD stays intact. Serialize scene mutation and render stepping through the runtime owner. Read references/ovstage-ovrtx-integration/README.md for that loop, then the population or data-plane references when implementing scene loading or attribute writes.

Keep expensive loading and metadata work off the UI and render threads. Keep transport details behind a delivery adapter so camera, selection, properties, and loading state can be reused across local and streamed shells.

Add Capabilities As Needed

Use references/routing.md to select camera/picking, scene management, render settings, cloud assets, simulation, editor tools, deployment, or CAE/CFD visualization. Read the selected capability's implementation flow and its linked contract; the full reference catalog is not a prerequisite for a first frame.

For simulation, use the routing reference's Simulation Ecosystem section plus references/realtime-sim-integration/README.md. For UI-driven forces or control, add references/interaction-round-trip/README.md.

For example scenes, use references/usd-sample-data/README.md. When a runtime failure needs diagnosis, use references/troubleshooting/README.md and the selected SDK's matching release guidance before changing the architecture.

Solid Modeling: Upstream First

For usd-brep, exact BRep, solid modeling, kernel failures, CAD topology, or BrepArray questions, first read references/dependencies/README.md → USD BRep. Follow the owning repository's current agent guidance, skills, operation docs, and package API before selecting local integration references. Discover the upstream router rather than assuming skill names or directory layouts remain fixed. Upstream owns modeling semantics, schema rules, tuning, and diagnostics; this package owns OVRTX display, worker handoff, selection mapping, and publication timing.

GPU Processing And Stylized Viewers

For custom GPU processing, chained outputs and lens sampling, read references/spg-processing/README.md. For toon, ink, technical/X-Ray styles, geometry strokes or full wireframe, continue to references/stylized-visualization/implementation.md. Its USD/Lua/CUDA starting node and geometry/depth/composite build sequence work with the selected viewer recipe; use references/stylized-visualization/style-recipes.md for the preset math and references/stylized-visualization/README.md for supporting contracts. Implement the result inside the existing OVRTX/runtime owner and deliver the published processed output. These instructions do not require a sample checkout or a prior stylized-rendering implementation.

For an optional usd-brep geometry source, first read references/dependencies/README.md → USD BRep and the owning repository's current agent guidance, skills, and operation docs; then read references/stylized-visualization/usd-brep-source.md for the display boundary.

XR Viewers

For headsets, immersive viewers, or spatial editors, start with references/xr-viewer-recipe/README.md. Resolve SDK setup and contracts through references/dependencies/ovxr.md; the library owns OpenXR lifecycle and frame submission. For controllers, hands, navigation, or scene tools, add references/xr-spatial-input/README.md and references/ovxr-python-interactions/README.md. For projection-integrated UI, read references/ovxr-composition-ui/README.md and references/ovxr-composition-ui-validation/README.md.

Files

references/ovui-local-viewer-recipe/interaction-features.md

ovui Local Interaction Features

7. Add Camera Controls

Read references/viewer-input-routing first for ovui button normalization, click-vs-drag dispatch, and input/render-loop ownership. Then implement camera math in local_app/input_controller.py and local_app/camera_controller.py:

  • Track mouse press, movement, release, buttons, wheel, keyboard modifiers, viewport dimensions, and drag threshold state from ovui callbacks.

  • Convert ovui screen coordinates into render-image coordinates, accounting for widget position, preserve-aspect scaling, and letterboxing.

  • Use drag thresholding so a short left press/release selects and a left drag orbits.

  • Support orbit, pan, dolly/zoom, wheel zoom, fit-to-stage, and optional WASD fly mode when requested.

  • If selected-prim transform is enabled, test transform-gizmo presses before normal left-drag orbit and before click selection.

  • Sanitize camera state before input handling and before writing matrices.

  • Enqueue the viewer camera's local omni:xform through the OVStage runtime adapter, then publish it before rendering the committed ordinal. Use direct renderer writes only for an explicitly standalone compatibility path.

Critical contracts:

  • The camera is a USD prim. Camera movement writes omni:xform on the viewer camera path.

  • Use the row-vector matrix layout expected by USD/ovrtx: basis vectors in rows and translation in the final row.

  • Clamp elevation away from straight up/down singularities.

  • Clamp camera distance above a small positive minimum.

  • Skip camera writes when any matrix value is non-finite.

  • Use the same world-up convention as the scene when fitting and orbiting.

  • Left-click selection fires only on release if movement stayed below the drag threshold.

  • A transform-gizmo mouse-down owns the full gesture until release; it suppresses camera orbit and click-pick for that mouse-down.

  • Map ovui button IDs to the camera helper's expected button IDs before interpreting gestures.

Decision points:

  • If the user wants simple navigation, use left drag orbit, middle drag pan, right drag dolly, and wheel zoom.

  • If the user wants DCC-style navigation, use Alt+left for orbit, Alt+middle for pan, and Alt+right for dolly while preserving left-click selection.

  • If the stage has an authored camera and the app policy is to use it, initialize viewer camera settings from that camera before allowing interaction.

  • If the user requests camera UI buttons, wire them to local runtime commands rather than duplicating pointer gesture logic in widgets.

  • If a minimal ovui gizmo renders handles but does not expose handle-drag callbacks, project the selected pivot into viewport space and implement a direct-manipulation fallback that converts mouse deltas into camera-plane world deltas.

Common failure modes:

  • Putting matrix basis vectors in columns instead of rows places the camera inside, behind, or under the scene.

  • Ignoring letterboxing makes orbit centers and picks offset from the visible image.

  • Treating every left release as selection causes accidental selections after orbit drags.

  • Letting the camera controller see a transform-gizmo press first makes the gizmo appear inert because orbit owns the drag.

  • Forgetting ovui's button order makes right-drag and middle-drag modes swap.

Read for depth: see references/viewer-input-routing, references/camera-controls, references/local-viewer, and references/stage-hierarchy for input routing, camera math, and bounds contracts.

8. Add Picking, Selection, And Highlighting

Read references/native-picking-selection, references/object-selection, and references/selection-feedback before implementing this section. Those references own the current native OVRTX picking and selection shapes; this recipe should route to those contracts instead of generating compatibility branches or fallback picker modules.

Do this in local_app/selection_controller.py; the completed click or marquee gesture should come from references/viewer-input-routing:

  • On left mouse release after a click gesture, map the screen coordinate to a render pixel and enqueue a native ovrtx pick query.

  • After the next render step, read the synthetic ovrtx_pick_hit render var, validate its params, resolve primPath ids to USD prim paths, and deduplicate the result.

  • Maintain selected tree path and selected mesh paths separately when an Xform or Scope expands to descendant geometry.

  • Keep selection state in runtime memory and update viewport highlight, tree row selection, and prim info from that single source.

  • Clear selection on scene reset, scene switch, empty tree selection, or explicit clear command.

  • Apply visual selection feedback with native renderer selection-group helpers, not by writing legacy attributes or permanently editing the user USD.

Use the current shapes from native-picking-selection; the essential contracts are a normalized synchronous enqueue, renderer-owned selection groups, and an automatic pick-hit output:

renderer.enqueue_pick_query(
    render_product_path=render_product_path,
    left_ndc=left / render_width,
    top_ndc=top / render_height,
    right_ndc=right / render_width,
    bottom_ndc=bottom / render_height,
)
path = renderer.resolve_prim_path_id(int(prim_path_id))
renderer.set_selection_group_styles({
    1: ovrtx.SelectionGroupStyle(
        outline_color=(1.0, 0.6, 0.0, 1.0),
        fill_color=(0.0, 0.0, 0.0, 0.0),
    )
})
renderer.set_selection_outline_group_strings(
    paths,
    groups_uint8.tolist(),
)

Decode the post-step synthetic pick render var with ovrtx.OVRTX_RENDER_VAR_PICK_HIT, validate magic, version, and hitCount, read the named primPath tensor, and resolve IDs before mutating selection. Do not scaffold GpuPicker, CPU ray/AABB fallback, segmentation ID maps, or Warp outline systems for new generated apps.

Critical contracts:

  • The pick query rectangle uses render-product pixel space after letterbox/scaling correction.

  • Convert those pixels to normalized top-left bounds for enqueue_pick_query(); do not add ovrtx_pick_hit to orderedVars because the query produces it.

  • Picking coordinates must use render-pixel space after letterbox and scaling correction.

  • Selection state is local-runtime authoritative. Widgets mirror it; they do not independently mutate it.

  • Enable selection outlines at renderer creation and set per-group outline/fill colors with Renderer.set_selection_group_styles(...).

  • Clear previous outlines with group 0; assign selected prims to a non-zero group such as 1 through set_selection_outline_group_strings().

  • Do not let selection picking run while the scene is loading or the renderer is resetting.

  • Check operation status for the stage load and render step before changing selection. enqueue_pick_query() registers synchronously; use the async form only when its returned operation is explicitly awaited.

Decision points:

  • If the user only asks for tree selection, implement tree-driven selection first and defer viewport picking.

  • If the user asks for visual highlight, use native selection outlines after selection state works.

  • If the user asks for hover, multi-select, or marquee selection, extend the local state model explicitly and keep final selected state centralized.

  • If selection needs property display, trigger the info panel to query the selected prim rather than stuffing full properties into every selection update.

Common failure modes:

  • Selection appears offset when coordinate transforms ignore image scaling or letterboxing.

  • No pick result arrives when the query was enqueued for a different RenderProduct than the next renderer.step().

  • Highlight persists across scene loads when old selection outline groups are not cleared.

  • Tree selection and viewport selection diverge when each widget tracks its own selected path.

Read for depth: see references/viewer-input-routing, references/native-picking-selection, references/object-selection, references/selection-feedback, references/prim-info-display, and references/stage-hierarchy for the full input, picking, highlighting, and info contracts.

9. Add Scene Switching And Asset Browsing

Put the scene registry and load state in local_app/scene_manager.py; the scene picker displays friendly labels and submits stable paths or asset IDs. Use references/stage-management for one preparation-and-handoff flow, and references/cloud-assets for remote preparation.

Keep the current scene interactive while resolving/downloading the next asset. Only the latest requested replacement may reach the runtime handoff. At that point, retire old scene queries/picks, populate a fresh Stage, and restore the chosen settings/camera policy before reporting its first valid frame. A failed handoff keeps the last image visible with an error state and a retry action.

Use the same replacement path for Reload. If the product needs to retain the old live scene until the new scene succeeds, budget for both generations rather than assuming reset provides rollback. Additive edits use the separate composition path in references/ovstage-population.

10. Add Stage Hierarchy, Properties, And Variants

Do this in local_app/stage_queries.py, local_app/widgets/stage_tree.py, and local_app/widgets/prim_info_panel.py:

  • Query the current stage through native runtime queries first, then through the chosen pxr subprocess mode for OpenUSD features that still require it.

  • Query root children after a scene loads.

  • Load tree children lazily when a row expands.

  • Represent each prim with name, path, type, and child-load state.

  • Fetch properties for the selected prim when selection changes.

  • Fetch variants for the selected prim and apply variant changes through the scene manager.

  • After variant changes, refresh affected hierarchy, properties, bounds, native pickability state, and selection if composition changed.

Critical contracts:

  • Avoid full recursive hierarchy traversal by default for large scenes.

  • Keep children semantics stable: expandable-but-not-loaded is distinct from loaded children and leaf rows.

  • Do not assume all USD property values are directly displayable. Normalize arrays, tokens, paths, numbers, booleans, and fallback display strings.

  • Include the prim path with every local query result so widgets can ignore stale data after selection or scene changes.

  • If query work can block the UI, move it to a worker and return results to the UI/render loop safely.

  • Tree selection should call the same runtime selection path as viewport picking.

Decision points:

  • Use direct pxr imports only after verifying the exact local process import path for the selected OVRTX/OVStage/USD wheels.

  • Otherwise, use a subprocess query mode and keep logs on stderr while requests and responses use structured data.

  • If the user asks for variant editing, treat it as a scene mutation and route it through scene management.

  • If the user asks for full property editing, treat it as a separate feature and add explicit edit/apply/reload contracts.

Common failure modes:

  • Large scenes freeze the UI when the full hierarchy is traversed synchronously.

  • Non-serializable or non-displayable USD values break property rendering.

  • Variant changes leave stale property and child rows unless caches are invalidated.

  • Direct pxr imports can conflict with ovrtx if import order or library paths are inconsistent.

Read for depth: see references/stage-hierarchy, references/prim-info-display, references/stage-management, and references/windows-native-setup for the full hierarchy and property contracts.

11. Add Render Settings And Lighting Controls

Do this in local_app/render_settings.py, local_app/settings_store.py, and local_app/widgets/render_settings_panel.py:

  • Define a small persistent settings model for validated render settings, camera policy, segmentation/debug state, viewport/profile defaults, and non-live defaults after the runtime capability list is known.

  • Build a runtime-owned supported-settings capability list from verified backend apply paths. The render settings panel must render from that list, not from hard-coded optimistic controls.

  • Load settings at app start.

  • Apply validated immediate settings and accepted profile/default settings after every scene load.

  • Persist user changes after validation.

  • Keep scene-independent settings separate from scene-specific transient state.

  • Echo the effective settings and capabilities back into UI state after every change so controls reflect clamped values, applied, applies_at, requires_reload, and any message.

Critical contracts:

  • Do not add viewer lights by default. Only create viewer-owned lights when the user requests lighting controls.

  • If changing resolution, treat it as a render product, image bridge, pick coordinate, and viewport math reconfiguration.

  • If changing render vars, update scene setup and frame extraction together.

  • Persist validated settings across scene switches.

  • Validate every setting value from the UI before applying it.

  • Reject unsupported setting keys. Do not report success for client-side form state alone.

  • Do not invent write_attribute names for renderer internals; expose only controls supported by the active ovrtx build and documented by references/render-settings.

Decision points:

  • If the user wants a basic Omniverse Realtime Viewer, provide only verified immediate controls by default. AOV/debug view is acceptable when implemented through aov-switching; lighting controls are acceptable only when backed by a verified live apply path or explicit reload/profile workflow.

  • If the user wants material or renderer internals, read references/render-settings before exposing them.

  • If the setting can be applied live through ovrtx attributes, apply it from the render loop.

  • If the setting requires stage reload, expose it as an explicit render-profile or scene-load action rather than a live control.

  • If the user wants pick-driven effects, runtime transforms, or physics impulses, expose them as separate optional capabilities and route accepted changes through the runtime owner. OVPhysX USD population should run in a worker unless PhysX.attach_ovstage(stage, read_ordinal=...) is verified for the exact installed OVStage/OVPhysX ABI.

Common failure modes:

  • UI says a setting changed while the renderer is still using the old value because effective settings were not echoed back.

  • UI exposes controls that cannot apply because the panel did not render from backend-advertised capabilities.

  • Viewer-created lights change the look of authored scenes unexpectedly.

  • Resolution changes break picking because render product, image bridge, pick coordinate mapping, and letterbox math no longer agree.

  • Extra render vars increase memory and frame time even when no feature uses them.

Read for depth: see references/render-settings, references/stage-management, and references/stage-loading for the full settings contract.

12. Manage Runtime State And Shutdown

Do this in local_app/runtime.py:

  • Maintain explicit runtime states: starting, idle/no scene, loading, rendering, error, and shutting down.

  • Drain queued commands from the UI/render loop before stepping the next frame.

  • Keep the latest loaded stage path, hierarchy root, selection, render settings, loading state, and viewer error in runtime memory.

  • Handle app shutdown by stopping the ovui loop, preventing new scene commands, releasing renderer-owned resources if the API exposes them, and closing any query subprocesses.

  • Guard long-running operations so the UI can display loading or error state instead of silently freezing.

Critical contracts:

  • UI callbacks must not call renderer load/reset/step/write APIs directly if they can run during another renderer operation.

  • Scene load and reset operations must be mutually exclusive with frame stepping.

  • Errors should clear active drag state, loading state, and stale selection when needed.

  • Query subprocesses must be terminated on shutdown and restarted only through a controlled path.

  • The app should exit cleanly on interrupt or window close without leaving stale GPU processes.

Decision points:

  • If all work stays on the UI thread, keep command handling simple and deterministic.

  • If loading large scenes needs a worker thread, make the render thread the only component that mutates the ovrtx renderer.

  • If hierarchy queries are slow, cache per-prim results and invalidate them on scene switch or variant change.

  • If the app needs autosave settings, write only app settings JSON, never user USD files.

Common failure modes:

  • Calling renderer methods from both callbacks and the render loop creates rare crashes that are hard to reproduce.

  • Exceptions during input can leave a drag gesture active and cause later clicks to orbit or select unexpectedly.

  • Query workers that write logs to stdout can corrupt structured responses.

  • App exits that skip GPU cleanup may leave stale Python processes holding CUDA/RTX state.

Read for depth: see references/local-viewer, references/ovrtx-rendering, references/stage-management, and references/stage-hierarchy for runtime and shutdown contracts.

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Needs: Needs Python, Needs Node.js

The author says: Orchestrator skill. Downstream focused references may require NVIDIA GPUs, ovrtx, ovstream, ovui, OpenUSD, Python, Node/React, Tauri, Electron, C++, or cloud GPU deployment access depending on the selected viewer path.

Security

What our scanner noticed· Medium risk

  • Large base64-encoded blobs
  • Instructions that use the network

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Created by
NVIDIA Omniverse
Publisher
NVIDIA (Official)
Source
NVIDIA/skills · skills/omniverse-realtime-viewer · commit fce3cca
Licence
Apache License 2.0 (redistribution allowed)
  • LICENSE (added by AutoGineer from LICENSE-APACHE)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by NVIDIA Omniverse or NVIDIA.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download omniverse-realtime-viewer.zip