← Back to timeline
NVDAKeynote2026-07-20

The speaker describes compressing a large general generative model into a smaller specialized model for real-time rendering.

Loading moment…

Edward Liu · NVIDIA · NVIDIA Keynote Live at SIGGRAPH 2026

So, we distilled. We took what those general model that knows about the world appearance and compressed it into a much smaller one-step pixel space diffusion transformer model that does only one thing. Make real-time rendering look more real.

Transcribed from the source video. Automatic captions, corrected for known mis-transcriptions.

Key terms

distillation
Training a smaller model to reproduce the behaviour of a much larger one so it runs faster and cheaper.

Educational summary of a public statement. Not investment advice.