NVIDIA DLSS 5 Mod Runs In A Browser And Even Works On Apple Silicon
Created by a developer known as MAAN and spotted by Videocardz, this web implementation of DLSS 5 is little more than a model viewer that has the ability to apply the DLSS 5 neural filter to loaded models. It offers a chance to preview the look of DLSS 5 on most platforms, as unlike the official release (currently only available in NBA 2K27), it does not require a GeForce RTX 50 series graphics card. I was able to get the demo working on my ASUS ROG Flow z13 tablet with Radeon 8060S graphics, although it didn't load on my aging Snapdragon 888-powered smartphone.
That could have been down to the amount of memory required. The models that the site includes are pretty big, but the DLSS 5 neural network itself is the largest amount of the download, totaling nearly 150 MB by itself. This is pretty large for a website, and some devices may balk at the bulk. However, the developer notes that it works on MacOS, and in theory it should work on most devices with a compatible browser.
If you keep up with DLSS news, it should come as no surprise that the page is able to run the DLSS 5 neural filter on nearly any hardware, as it's not doing anything unique to NVIDIA's hardware. Indeed, modders have gotten DLSS 5 working on older GeForce hardware, AMD and Intel hardware (including integrated graphics), and even on video feeds with no 3D information. While DLSS 5 is AI-powered, it's mostly a post-processing filter, so you can apply it to nearly anything, and it can be run on nearly anything with a little tweaking.
DLSS 5 running in the browser with #webgpu And yes it works on MacOS too. Try the the live demo here https://t.co/frqdwHzeDv You can also try it with your own models #WebDev #AI #threejs https://t.co/qYc0cvJLLu
— MAAN (@maanalaolaqy) September 16, 2026
As Whycry notes, the interesting part of MAAN's demo is how he's wired it up to the Three.js framework. NVIDIA doesn't offer WebGL or WebGPU as supported DLSS interfaces, which means the developer has either extracted the weights and reimplemented the network or created some kind of shim to use the leaked binary directly. In any case, it's impressive stuff, and you should check it out if you have a minute.

