Unity Neural slashes graphics memory in URP
Unity announced Unity Neural, embedding native neural inference into the Universal Render Pipeline (URP) compute architecture. Powered by Unity Sentis, it introduces in-editor Neural Texture Compression (NTC) to slash runtime VRAM and package sizes using custom-trained models.
Embedding neural inference directly into the engine's compute pipeline shows how AI can solve practical hardware bottlenecks instead of chasing generative gimmicks. Shifting from static block compression formats to lightweight on-device neural decoders makes high-fidelity visual assets viable across memory-constrained mobile and XR platforms.
- –Cuts runtime texture memory by up to 50% and asset package sizes by roughly 70% while preserving fine detail.
- –Executes on-device via Unity Sentis with zero cloud dependencies, eliminating latency overhead for interactive rendering.
- –Enables developers to train bespoke compression models directly inside the Unity Editor on project assets.
- –Strengthens Unity's push toward a unified URP architecture by pairing neural decompression with modern upscalers.
DISCOVERED
9m ago
2026-11-20
PUBLISHED
58m ago
2026-11-20
RELEVANCE
AUTHOR
unitygames