Generate neural graphics datasets with Neural Graphics Data Capture in Unreal Engine
Introduction
Benefits of Neural Graphics Data Capture for game developers
Install and enable the plugin
Configure Level Blueprint capture controls
Run capture and verify outputs
Capture settings and troubleshooting
Next Steps

  Log an issue

  Fork and edit

  Discuss on Discord

Available capture settings

You can tune these in NGDCRenderingSettings and NGDCExportSettings:

  • UpscalingRatio: Sets the ratio between jittered input and ground-truth output. Fractional values are supported.
  • SupersamplingRatio: Controls quality/performance tradeoff while capturing.
  • FixedFrameRate: Locks capture frame rate when greater than 0.
  • CameraCutTranslationThreshold and CameraCutRotationThresholdDegrees: Heuristics for marking camera cuts when needed.
  • DatasetDir and CaptureName: Control output folder location and capture name.

For smooth out-of-box capture, a representative spec is:

  • RAM: 64 GB
  • GPU: NVIDIA GeForce RTX 4080 (driver 572.96 or newer)
  • OS: 64-bit Windows or Linux

Common issues and solutions

Plugin not detected:

  • Confirm plugin is under your project Plugins/ directory.
  • Regenerate project files and rebuild.

Hotkeys don’t work:

  • Verify Level Blueprint has C wired to Begin Capture and V wired to End Capture.

Dataset folder is empty:

  • Confirm capture started and that export path is writable.

Stutter or instability:

  • Lower SupersamplingRatio and test again.

Unexpected output size:

  • Use Standalone Game mode instead of New Editor Window (PIE).

What you’ve learned

You’ve successfully set up a workflow to capture neural graphics datasets directly from Unreal Engine 5.5 gameplay. You can now:

  • Generate training data from your own game content without graphics API configuration
  • Capture representative gameplay sequences with real motion and camera behavior
  • Export frame datasets ready for use with Model Gym’s NSS data generation pipeline
  • Tune capture settings to match your specific quality and performance requirements

Use this repeatable workflow for model experimentation and evaluation in your neural graphics pipeline.

Back
Next