Today we are going to breaking down one of the most important legal decisions taken by the court in AI because nothing brings people together like copyright fights and a giant , pile of digital confusion.
So what exactly happened?
Getty Images, the giant photo library got very angry at Stability AI, ah, that folks behind Stable Diffusion.
Getty Images said Stability AI trained 12 millions of their photos without the permission of them and used them to create a , new AI data set.
So Getty Images sued Stability AI for intellectual property theft.
But Stability AI replied with something like no, no, we didn’t copy your images.
Our model only learned the patterns. The AI doesn’t store your photos, it just remembers how to draw things vaguely kind of. It’s like someone who says, who read your diaries, burns it and they says it’s fine, I only memorized your emotions. The argument went for months.
During that time, Getty Images dropped down m several claims against Stability AI, such as the AI model itself trained inside the UK. But Stability AI used Amazon’s server. It was actually situated at outside the UK. So they dropped their claim. Last week the case came to a conclusion.
So the judge asked the main question, ‘is Stability Diffusion basically a giant folder of Getty Images photos disguised as a machine learning model? Getty Images says yes and Stability said no.
So the court started looking at the numbers.
The dataset Getty complained about was nearly 220 terabytes for the 12 million images.
But the stability AI model weights around 3.4 gigabytes.
So the judge clearly said the model doesn’t store or contain all these images. It only learns the pattern from them.
Which is a basically legal way of saying the model is not a photocopier, it doesn’t include all your images. This was a huge deal because many people assumed AI models those copies inside them. Turns out they do not.
They just learn how to imitate , style structures like an art student, but faster and with less sleep. But even though Getty lost a big copyright point, they won a tiny piece. Some AI generated images from old version of Stability Diffusion, accidentally recreated a , Getty or Ice Dog watermark. Imagine AI saying I made this from scratch but ignore the Getty logo floating in the skyline. The court got that and that is a, ah, trademark infringement.
So Getty scored a small win there. But overall most of the trademark complaints are also flayed. The judge basically said these watermarks only showed up in the older models, not consistently. So why this decision matters? This case is one of the biggest legal milestones for AI so far.
It confirms something important. Training AI on copyrighted data is not same as storing on copyright on original works. The model learns ideas not storing the raw images.
From AI art tools to film production to VFX overflows to the friends who keep posting AI selfies like they’re staring a perfume ad. It also means company building AI tools cannot say we’re just a neutral tool.
Whatever comes out is the user problem. The judge said the developer has sole responsibility for training the pipeline and the output. So this ruling does not mean AI companies can scrap whatever they want. It doesn’t mean copyright is irrelevant now. It means the future lawsuits will get even more confusing.
Except my style was stolen, the AI copied my work, the AI output looks my dog’s Instagram picture from 2017. The legal world is trying to keep up with AI that the AI is trying to keep up with itself and everyone watching it like that.
