Someone on Twitter posted a Monet painting, claimed an AI made it, and asked the internet to describe everything wrong with it. And hundreds did. The reflections were just splattered noise. There was no depth or cohesion. The background, as one reply put it, was "an egregiously vague algae amalgam." No focal point. The further back you went, the less anything looked like anything.
The image was a real Monet. It's one of his late Water Lilies. Every flaw they listed is exactly what makes those paintings hang in museums. He spent the last twelve years of his life trying to kill the horizon line and dissolve the focal point. He painted on canvases big enough to wrap around you. The "mess" was the entire point. One of these paintings sold at Sotheby's in 2014 for $54 million. Monet gave eight more to France as a peace offering the day after the Armistice. They hang in two oval rooms in Paris that he and his architect designed themselves.
There's a name for what just happened. In 2023, three researchers at Columbia Business School ran an experiment with almost three thousand people. They showed each person the exact same painting twice, once labeled human-made, once labeled AI-made. People valued the AI-labeled version 62% lower and thought it took 77% less time to make. Same canvas, opposite verdicts.
A team at University College London had already shown why this happens. Back in 2009, they put people in a brain scanner and showed them art. Each painting appeared twice. One version was tagged "from a gallery," the other "computer generated." The reward-and-pleasure part of the brain lit up more strongly for the gallery tag. The label was changing how the brain saw the same image.
Knowing about the bias doesn't help. Over 70% of participants in the Columbia study said the paintings looked the same to them. They still rated the AI-labeled ones lower.
In 2023 a lab at Keio University in Japan tested how accurately people could spot AI art. People got human paintings right only 68% of the time. For AI paintings, the rate was 43%, worse than flipping a coin. A blogger named Scott Alexander ran the same test publicly in 2024 with fifty paintings, museum pieces mixed in with AI ones. The median score was 60%, barely better than guessing. Art professors and curators took the test. Most landed in the same range as everyone else.
The "splattered noise" the comments mocked is some of the most-studied brushwork in art history. The "vague blob" is on the walls of a Paris museum. The brain seems to stop seeing the painting the moment it's told a machine made it.
@starks_arq Your content barely ever makes sense. In this example they're getting shot at from behind, but then they look forward and shoot forwards - then the guy throws the grenade in front of him but the explosion happens behind him and the dudes in front just somehow disappear ๐
@JSFILMZ0412@SteveEricsson1 I signed up for business annual but all my generations failed overnight. When I upload character sheet to dreamina it works fine, but if I upload the same character sheet to topview, it fails?
@JSFILMZ0412@VahnAeris Whats the generation time atm - still 4 hours or so? Do you think it will get worse? I want to pull the trigger with the annual topview sub but if too many people sign up, the delay could be 8 hours + or worse? How much longer will they offer this deal I wonder