| name | ovrtx-current-scene-workflow |
| description | Build, edit, preview, and render the current Blender scene through the OVRTX Blender add-on. Use when a user wants OVRTX output from a new or existing `.blend`, wants to drive Blender through MCP and see live viewport updates, or needs a reproducible still from the current scene. The workflow uses the add-on's Blender boundary and an installed runtime |
| license | Apache-2.0 |
| metadata | {"author":"Max Bickley","version":"0.1","team":"omniverse","domain":"physical-ai","tags":["blender","omniverse","ovrtx","workflow"]} |
OVRTX current-scene workflow
The current Blender scene is the authoring boundary. It may be saved or unsaved,
created through Blender's UI, imported from USD, or assembled by Blender MCP.
OVRTX builds a managed scene generation from that state; use the documented
current-scene workflow rather than pre-exporting a separate USD.
When to Use
Use when a user wants OVRTX output from a new or existing .blend, wants to drive Blender through MCP and see live viewport updates, or needs a reproducible still from the current scene. The workflow uses the add-on's Blender boundary and an installed runtime.
Instructions
- Use
blender-mcp-setup to verify the Blender control loop, then use the normal Blender MCP tools or UI to inspect the scene.
- Save a working copy when the scene is valuable or edits are destructive. Record the
.blend path, active camera, resolution, frame, world, lights, and renderable objects.
- Set the scene camera explicitly. When named subjects must be wholly visible,
run
blender-camera-framing at the final resolution and require its safe-UV
and clipping checks to pass. Otherwise check clipping, lens/orthographic
scale, transform, resolution, and aspect ratio before debugging OVRTX.
- For imported USD, retain the source path and import settings. For a generated scene, use stable object/material/light names so later edits can be resolved.
- Do not delete existing objects or reset the world unless the user requested a clean rebuild.
2. Select OVRTX and configure a first render
After add-on preflight passes, read
references/blender-ovrtx-api.md and run
scripts/probe_ovrtx_scene.py. Set the scene render engine to
OVRTX_EXAMPLE only when the probe confirms registration. The add-on settings
are available as scene.ovrtx_example:
min_samples: first progressive sample count (use 1 for a smoke test);
max_samples: completion target (raise after the first valid frame);
color_presentation_mode: scene_linear_hdr when the consumer owns display conversion, or ldr_rgba8_display_passthrough when OVRTX returns display-encoded pixels;
sync_viewport_camera: map Blender viewport orbit/pan/zoom to the OVRTX preview camera when the scene camera mapping is valid.
Use Blender's Render Image/Render Animation action or the MCP's Blender-Python execution to request the frame.
3. Validate the first frame
Before increasing quality, verify all of the following:
- the OVRTX render engine is still active (not a silent Cycles/Eevee fallback);
- the expected camera is bound and the subject is in frame;
- the output dimensions and orientation are correct;
- the image is not black, flat, mirrored, or dominated by a missing world/material;
- the requested samples advance and the output file exists at the user-selected path.
Use a viewport screenshot plus structured Blender scene inspection. Keep raw OVRTX output separate from any presentation-grade crop or color grade.
4. Interactive refinement
In a rendered 3D View, enable camera synchronization and orbit/pan/zoom normally. The add-on reuses a warm OVRTX session and progressively refines stable views. If the viewport is stale, use the add-on's Restart Viewport Session action, then wait for a new first sample before judging quality.
Supported value edits (including transforms, camera view, and the add-on's mapped light/material/world values) should be made through Blender and allowed to propagate in-session. Structural or topology changes can request a new scene generation; wait for the replacement before comparing images. Do not edit the generated USD overlay by hand.
5. Color and output ownership
Choose exactly one display-transform owner:
- With scene-linear HDR, preserve linear values and apply Blender's intended view transform once in the consumer or presentation step.
- With LDR passthrough, treat OVRTX pixels as already display encoded and do not apply a second Blender transform.
Record render mode, sample range, resolution, camera, color mode, output path, and whether the frame is raw or graded. A visually pleasing screenshot without this provenance is not a reproducible render.
Recovery
- Black/empty frame: check camera binding, runtime status, world/light inputs, and session logs; retry at one sample.
- Wrong view: reset the viewport camera, make the intended camera active, and confirm
sync_viewport_camera and projection.
- Stale edits: restart the viewport session after a structural change; do not compare a cached frame with a newly authored scene.
- No render engine: return to
ovrtx-addon-install-and-preflight.
- MCP disconnected: return to
blender-mcp-setup; do not diagnose an OVRTX worker as an MCP failure.
Closeout
Summarize whether the requested native operation succeeded and the first
actionable blocker or approximation. Include camera, engine, sampling, color,
paths, and session details only when they help reproduce or review the result.
Always distinguish OVRTX-rendered artifacts from Blender fallback renders and
postprocessed images.