Mark Linington has spent over 20 years building game worlds. Mass Effect, Halo, Far Cry, Overwatch 2, Diablo IV. He knows what it takes to craft a specific look and feel. So when Nvidia showed off DLSS 5 and its AI-assisted rendering, he paid attention, especially considering the recent controversy over ownership claims for demonstration footage. And what he saw worried him. "The way the current demos have been presented, they have pushed the results into reinterpretation territory," Linington told Notebookcheck. That's a problem when artists spend hours dialing in precise styles. Push too hard and "that balance can be erased and, with it, the soul of the art behind it."
Nvidia says artists keep control through masking, intensity sliders, and color grading. Linington doesn't buy it. Those tools contribute "only a very small part of the overall look." He points to the "nano banana 2" workflow as a better model, one where artists feed exact reference images, color values, and lighting direction to the AI. That's hands-on. DLSS 5, from what he's seen, offers nothing like that level of control. There's also the photorealism problem. If the tech is tuned for realistic lighting and materials, it creates a built-in bias. Linington would "discourage any use of it on anything stylized or painterly, as that would likely erase all the nuance." Games that deliberately choose rougher or artificial aesthetics could lose what makes them distinctive.
He says DLSS 5 can't just be flipped on at the end of production. That gives poor results. It needs to be baked into the development cycle, tested and tuned alongside artists from the start. That means building test scenes early, comparing AI-enhanced output against the target look at every milestone, and treating the upscaler as another tool in the pipeline rather than a post-processing afterthought.
Linington confirms what many suspect: "Most, if not all, major game studios are using AI in their production pipeline right now, without a doubt." AI itself doesn't worry him. Careless use does. A one-click texture slapped onto a model and called finished is bad practice. This parallels concerns that AI agents risk generating output without genuine understanding. AI as a production partner that helps you work across art, animation, rigging, and sound? That excites him.