The Accidental Renaissance: A History of Early AI Pictures

If you scroll through a modern AI art generator today, you can conjure a hyper-realistic cinematic portrait or a surreal digital painting in mere seconds. But just a few years ago, the landscape looked vastly different.

The era of early AI pictures—spanning roughly from the mid-2010s to around 2020—was a wild frontier. It was a time of pixelated hallucinations, bizarre artifacts, and uncanny valley creations that felt less like polished digital art and more like transmissions from an alien dimension. Yet, this awkward phase laid the technical and cultural groundwork for the visual revolution we live in today.

1. The Dream Machines: DeepDream and Neural Style Transfer (2015–2016)

The public’s collective obsession with machine-generated art truly began in 2015 with Google’s release of DeepDream.

Originally built to help computer scientists understand how convolutional neural networks “saw” images, DeepDream took existing photos and over-processed them by enhancing patterns the AI recognized. The result? Ordinary landscapes and portraits transformed into psychedelic, nightmarish hallucinations packed with repeating dog faces, swirling eyes, and surreal geometric patterns. It wasn’t “creation” from scratch, but it was the first time machines showed us a distinct, recognizable aesthetic of their own.

Shortly after, Neural Style Transfer (pioneered by Gatys, Ecker, and Bethge) hit the scene. This algorithm allowed users to take any photo and redraw it in the artistic style of van Gogh, Picasso, or Munch. For the first time, algorithms could decouple a photo’s content from its style, giving everyday internet users a playful taste of machine-rendered creativity.

2. The Era of GANs and Pixels: When AI Learned to Draw (2014–2020)

While DeepDream and Style Transfer rearranged existing pixels, the real leap toward true generation came with Generative Adversarial Networks (GANs), introduced by Ian Goodfellow in 2014.