Overview
- A developer known as MAAN posted a live WebGPU demo that routes NVIDIA’s DLSS 5 model to render browser 3D content and said the project source will be released on GitHub.
- NVIDIA’s DLSS SDK and Streamline/NGX are built for native DirectX and Vulkan apps that supply motion vectors, depth and frame data, and NVIDIA’s documentation does not list WebGL or WebGPU as supported inputs.
- MAAN has not published technical details, so it is unclear how browser rendering resources are fed into DLSS 5; possible methods include a native bridge or external process but the author has not confirmed any approach.
- Performance remains the main limit: the demo is slow to load the neural model (reported about one to two seconds even on an RTX 4090) and runs sluggishly on macOS, making it a proof of concept rather than production-ready for real-time gaming.
- If MAAN publishes the code, the work could point to new ways to add native AI upscalers to web apps for visualization and previews, but widespread use will require solutions for motion/depth data, latency and cross-platform support.