GAN (generative adversarial network)
Two neural networks compete: one creates fakes, the other tries to expose them.
The generator produces images; the discriminator judges whether they are real or artificial. Through this competition the generated images become ever more realistic. GANs were introduced in 2014.
GANs shaped the first wave of photorealistic AI faces and deepfakes. For image generation they have since largely been replaced by diffusion models.
General information, not legal or professional advice.