Watch the NFRU technical deep dive
Watch the technical explanation of Neural Frame Rate Upscaling (NFRU) and its place in a mobile rendering pipeline:
Follow a frame through NFRU
NFRU generates an intermediate frame between neighboring rendered frames. It uses three major stages:
- Estimate motion: Optical flow estimates dense pixel-level movement between rendered frames. Engine motion vectors provide additional information about projected scene movement.
- Generate the frame: NFRU aligns candidates from neighboring frames. A neural network combines them and resolves missing or conflicting information.
- Pace presentation: Rendered and generated frames need to reach the display at an even cadence. You can use the SDK component or own frame pacing in your engine.
Your engine supplies color, depth, and motion vectors. It also supplies camera data and metadata. The SDK API workflow provides the exact dispatch sequence.
Understand where the headroom comes from
NFRU still needs a stable base cadence and adds processing and latency. It creates value when interpolation costs less than rendering another full frame. Choose one goal before testing: a higher presentation rate or more headroom for sustained performance and visual features.
Own the integration boundaries
Your game owns the following behavior:
- Input correctness
- Temporal resets
- Presentation
- Content testing
- Fallback behavior
Use the NFRU integration checklist as the implementation reference.
What you’ve learned and what’s next
You’ve learned how NFRU combines motion estimation, neural composition, and frame pacing.
Next, you’ll test the complete path against your content.