Nvidia researchers show how far AI image generation has advanced in recent years in creating realistic and customizable faces
Those people on the right aren't real; they're the product of machine learning — Developments in artificial intelligence move at a startling pace — so much so that it's often difficult to keep track.
Context & Ripple Effects
This demo extends work Nvidia researchers had already published: earlier in 2018 they showed how models trained on celebrity photos could produce faces that appear real but aren't by detecting patterns in training data, and a 2017 research effort had already turned flat portraits into 3D face images. What changes here is customizability — the output is no longer just plausible, it is controllable.
The significance shows up in what follows in the corpus: within two years, startups were selling computer-generated faces to clients like dating apps for advertising, and by late 2025 image generators were chasing realism by imitating phone-camera traits like exposure and sharpening. This 2018 result is an early marker on that curve.
First-order effects
- Nvidia cements its position as the visible research brand in generative imagery, with its lab output now indistinguishable-from-real faces rather than obviously synthetic ones.
- Editors, platforms, and readers lose the default assumption that a photographic face depicts a real person — verification burden shifts to whoever publishes the image.
Second-order effects
- A commercial market forms around the capability: per the corpus, AI startups begin selling realistic synthetic faces to advertisers and dating apps that want diverse imagery without hiring models.
- Rival labs and image-tool makers must match both realism and control; the later corpus evidence of generators mimicking phone-camera artifacts shows realism itself becoming the competitive axis.
Third-order effects
- If the pattern holds, synthetic human imagery becomes a commodity input for marketing and media, and the scarce asset moves from generating faces to proving which images are authentic — a structural problem the corpus shows persisting years later.
- The same generative pipeline broadens beyond stills into interactive synthetic humans, as seen in Nvidia and Convai's generative-AI game NPCs, pointing toward generated people across media formats.
The trend: Synthetic human imagery is moving from research demonstration to commercial commodity, with each realism advance shrinking the gap between generated and photographed faces.