Table of Contents
Nvidia provided more DLSS 5 footage during its keynote at SIGGRAPH 2026 yesterday. We’re getting ever closer to the DLSS 5 release date this fall, and Nvidia has been pretty silent about the project following its initial reveal at GTC in March – until now. The neural rendering visuals faced immediate backlash from gamers, who were quick to label it an “AI filter” that disrupts the art.
During the presentation, Nvidia’s Edward Liu went into further detail about how the technology works under the hood. He says, “DLSS must change the image in a way that it does not change the story” – referring to how DLSS 5’s neural rendering is trained to preserve artistic intent while maintaining a consistent image. Developers may fine-tune the visuals with a set of sliders.
DLSS 5 is designed to preserve artistic intent #
“I think it looks great. It’s a new level of realism, but looking real here is only half the problem. The other half is staying true to the original content, and that is tricky because all the generative models are probabilistics.”
“We’re going to use the frame the game just rendered. It’s not a prompt, it’s not a reference, it’s not a proxy scene, it’s a render’s own output. And it’s one signal that is actually universally available across all the engines. Unlike a prompt, this is actually a very information-dense and pixel-aligned view of the current shot… so we took the render frame and make it more realistic.”
Edward Liu, Director of Applied Deep Learning Reserarch at Nvidia
During the presentation, Liu explained three “fundamental and tough research challenges” of developing DLSS 5 and ensuring its visual makeover stays true to what is originally rendered by the game engine. “We draw a hard line and train the model in a way to respect intent,” says Liu, using aspects such as albedos, surface normals, and lighting information to restrain and guide the model. He makes it clear that
Latest PC & Tech deals
Acer Predator Helios Neo 16 - was $2,199 now $1,749
ASUS SFF-Ready RTX 5080 - was $1,499 now $1,354
Samsung Odyssey G7- was $999 now $649
HP OMEN 45L gaming PC - was $2,499 now $2,309
BIWIN Black Opal NV7400 4TB SSD - was $599 now $459
Prices correct as of July 10th, 2026.
- Challenge 1 – Preserving Artistic Intent
- Challenge 2 – Frame By Frame Streaming
- Challenge 3 – Real Time 4K
Nvidia displayed several slides showcasing the DLSS 5 developer tools, which provide game devs with access to global controls such as ‘Structure Intensity’ and ‘Tone Intensity’ with a slider from 0-1. There’s also a ‘Developer Masking’ feature which allows artists to manipulate objects individually – automatic masking identifies models or characters for devs to manually tinker with.
As such, developers can hand-pick what they want to alter with DLSS 5 – and by how much. This likely comes as a relief for gamers quick to judge the ‘AI filter’ altering how characters look. Resident Evil Requeim’s Grace Ashcroft was a leading example in Nvidia’s original demonstration earlier this year. Devs could simply choose to ignore DLSS 5 for character models.
Three models will be shipping with DLSS – Model A, Model B, and Model C – trained with different parameters and with a different interpretation of a rendered scene/frame. Developers can mix and match which model is used throughout the game.
Overall, it’s nice to see more behind-the-scenes on DLSS 5 ahead of launch. It certainly seems that Nvidia aims to silence the backlash by showcasing its versatility. DLSS 5 is far from a filter, and it gives artists full control over what is actually being shown in the final image. One thing we’re waiting for is more on its performance hit – the original March preview required two RTX 5090s, but it is being optimized to run on a single GPU; the new neural rendering feature will be exclusive to the RTX 50 series.