Enhance, Don't Change: DLSS 5, the Claims, and the Canary
On March 16, Nvidia unveiled DLSS 5 at GTC 2026. Jensen Huang called it the company's "most significant breakthrough in computer graphics since the debut of real-time ray tracing." Within 24 hours, the gaming internet had a different name for it: AI slop.
The gap between those two assessments is worth a closer look.
The Announcement
DLSS 5 — Deep Learning Super Sampling in its fifth iteration — is a departure from previous generations. DLSS 1 through 4 were fundamentally performance technologies: AI upscaling lower-resolution frames to higher resolutions, generating additional frames to boost fidelity and smoothness. They made games run better. Players noticed, liked it, and adopted it. By DLSS 4, over 750 games supported some version of the technology.
DLSS 5 is categorically different. It is a real-time neural rendering model that takes the game's rendered frame and motion vectors as input, then uses an AI model to infer and generate photoreal lighting, skin, fabric, hair detail, and material properties. It does not upscale. It does not generate frames. It repaints. It filters.
Nvidia's marketing described the technology as "3D-conditioned" and "anchored to source 3D content" — language implying that the model understands the geometry of the scene it's enhancing. In a subsequent email exchange between Nvidia's Jacob Freeman and YouTuber Daniel Owen, that claim was clarified. DLSS 5 takes a 2D rendered frame plus motion vectors as its sole input. It does not read geometry, depth buffers, or material properties from the game engine. Materials are inferred from the player's perspective alone.
Jensen Huang, appearing on the Lex Fridman podcast days later, described DLSS 5 as "3D-conditioned, 3D-guided." His own technical representative had already confirmed otherwise.
The Grace Ashcroft Problem
The showcase games included Starfield, Hogwarts Legacy, Assassin's Creed Shadows — and Resident Evil Requiem, Capcom's most recent tentpole release and critical success. The before-and-after images of Grace Ashcroft, a central character in the game, became the story.
With DLSS 5 applied, Ashcroft's facial features were visibly altered. Skin appeared smoother, facial structure subtly different, makeup-like effects generated where none existed in the original model, hair detail added in areas where the original geometry had none. The model had looked at a 2D frame, inferred what a "photorealistic" version of that face should look like based on its training data, and generated accordingly. The result was a different person.
This is not a minor artefact complaint. Resident Evil is a franchise whose players have followed specific characters across multiple titles. Altering a character's face through a rendering layer — without developer approval, as part of an Nvidia showcase — was always going to generate exactly the backlash it received.
Capcom developers who spoke to Insider Gaming described finding the renderings of Grace shocking. More significantly, Capcom was reportedly blindsided by Nvidia's use of Resident Evil Requiem in the presentation and feared the game would be associated with AI. The company's executive producer Jun Takeuchi had provided a supportive quote for Nvidia's press release. The people making the game had a different reaction.
The Response Arc
Jensen Huang's initial response at a GTC press Q&A with Tom's Hardware was unambiguous. Critics were, he said, "completely wrong." DLSS 5 was "not post-processing at the frame level." Developers retained full artistic control.
Within a week, the position had softened considerably. On the Lex Fridman podcast, Huang said: "I think their perspective makes sense, and I can see where they're coming from, because I don't love AI slop myself. All of the AI-generated content increasingly looks similar, and they're all beautiful."
That is a real concession. Huang was acknowledging the homogeneity problem — that AI-generated content converges toward a particular aesthetic of beauty that is recognisable as artificial precisely because it is optimised rather than authored. He was empathising with the criticism while defending the technology. The tension between those two positions was left unresolved.
Capcom resolved its own position more cleanly. In a financial Q&A published March 23, the company stated: "Our company will not be implementing any AI-generated assets into our video game content. On the other hand, going forward, we plan to actively utilise this technology in order to improve efficiency and productivity of game development."
The line Capcom drew — AI for workflow, not for final assets — is the clearest articulation yet of what responsible AI adoption looks like in a major studio. It is also a direct repudiation of what the DLSS 5 showcase implied.
What DLSS 5 Actually Is
Huang's Lex Fridman comments introduced a detail worth examining carefully. He described a future state in which developers could "prompt" DLSS 5 — providing stylistic direction or a reference aesthetic — and the model would generate output "all with the intent of the artist." The word prompt is not incidental. It is the operational vocabulary of generative AI.
Elsewhere in the same interview Huang called DLSS 5 "content-control generative AI" — a phrase that tries to distinguish it from uncontrolled generation but in the process confirms the generative mechanism. Every single frame, he said, DLSS 5 "enhances, doesn't change." The distinction between enhance and change is doing a great deal of work in a sentence where the technology has demonstrably altered a character's face.
Heraclitus observed that you cannot step into the same river twice. A frame enhanced by an AI model inferring hair, skin texture, and material properties that did not exist in the source geometry is not the same frame. The semantic distinction collapses the moment you look at Grace Ashcroft before and after.
The developer control available at launch — intensity sliders, colour grading, masking — does not extend to correcting specific generated details. Beyond reducing or disabling the effect, there is no mechanism to tell the model that a particular character should not have hair where the original geometry has none.
What It Means For Gaming Equities
DLSS 5 is not yet released — it launches in autumn 2026, confirmed for RTX 50-series cards only, with other architectures unconfirmed. That restricted compatibility matters. The RTX 50-series install base remains small, supply-constrained, and expensive. Developers integrating DLSS 5 at launch absorb implementation cost for a feature the majority of their players cannot access.
For names in the coverage universe, the implications split cleanly.
In spite of their flagship release's unwilling association with DLSS 5, Capcom has handled this better than most. The public separation between AI-for-workflow and AI-for-assets is credible, clearly communicated to investors, and — given Capcom's track record on capital discipline and brand management — likely to hold. The awkwardness of the Nvidia showcase was reputational noise, not signal. The franchise value of Resident Evil is not diminished by DLSS 5. If anything, Capcom's response reinforces the studio's commitment to the creative integrity that underpins its long-tail catalogue monetisation thesis.
CD Projekt Red is the more interesting watch. Cyberpunk 2077's extended rehabilitation depended almost entirely on restoring player trust through demonstrable quality. Any perception that DLSS 5 is altering the visual identity of a future title — without clear communication — carries asymmetric reputational risk for a studio still rebuilding goodwill. CDPR has not yet made a public statement on DLSS 5. That silence is worth monitoring as Witcher 4 development commentary emerges.
Nvidia itself — not a gaming equity in the traditional sense, but relevant to the coverage thesis — is structurally insulated from the backlash. With approximately 92% discrete GPU market share, DLSS adoption is effectively a developer requirement for premium PC optimisation. The DLSS 5 controversy does not threaten that position. What it does reveal is a tension in Nvidia's gaming narrative: the company is moving its developer-facing technology toward generative AI at a moment when the gaming audience is actively hostile to generative AI. Managing that tension, without alienating the developer relationships that sustain the DLSS ecosystem, is the company's primary communications challenge for the next twelve months.
The hardware oligopoly that taxes gaming software margins does not become less powerful because players disliked a character's face. But DLSS 5 makes visible something that was previously abstract: the direction of travel for Nvidia's gaming technology is generative, the developer "choice" to opt in or out operates under significant structural pressure, and the creative consequences of that pressure are now impossible to dismiss as theoretical.
Grace Ashcroft's face was the canary.
