How these photos are made

None of the people on this site exist. Every photo is created by artificial intelligence from a written description, and then goes through an automatic review before it is published.

Generative Adversarial Networks (GANs)

GANs were introduced in 2014. They work with two networks that compete: one generates images and the other tries to tell whether they are fake. After millions of rounds, the first one learns to create very convincing faces. That is how the sites that show “a person who doesn’t exist” every time you reload the page work.

GANs have limits: they usually make only front-facing faces and leave typical flaws in the background, earrings, glasses or teeth. They also can’t be asked for a specific scene, such as “a dentist treating a patient”.

Diffusion models

Today’s models start from an image full of noise and clean it up step by step, guided by the description. That is why they can create complete scenes: a waiter serving at a restaurant, a team meeting, the inside of a bakery. With good training data, skin, light and objects look natural.

How we work

1. AI agents write descriptions of useful scenes for each industry: website hero, team, service, premises, products and backgrounds for text.
2. An image model creates the photo on our own computers.
3. Another agent reviews it: it discards deformed faces or hands, strange text, brands and anyone who could look under 18.
4. The photos that pass the review are published with their title, alt text and keywords.

Every photo carries the AI-generated content label in its data and can be used for free under the license.