When X, formerly Twitter, announced its own take on generative AI images, it did so with seemingly few guidelines in place. Now, X’s Grok image generator is no longer hidden behind a paywall, making a limited version free for all X users. The move makes the AI that’s willing to generate images of recognizable celebrities and copyrighted art free to anyone with a Twitter account – and as a photographer, I find that utterly terrifying.
Last weekend X dropped the paywall behind Grok, opening up the AI to all users on the social platform. While paid subscribers still have the most features, all users can now generate up to four images per day and ask the chatbot up to ten questions every two hours. The only requirement to access Grok for free is an X account that is at least seven days old and linked to a phone number.
The platform also seemingly launched a new AI designed for creating realistic images, but then promptly removed the tool called Aurora.
Grok is notably different from other AI image generators because X calls it a system "with twist of humor and dash of rebellion." I tried out Grok’s AI generator back when I had to fork over $10 to Elon Musk, just to see if the AI really was as ethically lax as everyone claimed. As a journalist and photographer, I wanted to become acquainted with what the AI could do. I was dismayed when Grok spat out images that others like DALL•E and Google’s Gemini refused to generate.
Grok’s ability to generate images of recognizable celebrities and politicians means the AI can be used to create images perpetuating fake news. The AI seems to have gained a few guidelines since the initial launch flooded X with fake images of politicians wielding guns, later refusing to generate violent images with recognizable figures.
But the AI’s seemingly few guidelines also extends to intellectual property. Several AI image generators, including DALL•E and Google Gemini, will refuse to generate images in the style of a certain artist (though instead they may describe the artist's style and then generate an image in that style). Both of those chatbots also refuse to generate an image with a logo in them.
Meanwhile, when I tested Grok earlier this year, it included the CNN logo in a presidential debate image, even though I didn’t ask for such a background. I was also able to ask for images in the style of certain painters, and even name a specific work of art to replicate.
The one feature that appears to have changed since I first tested Grok earlier this year is that the images the free version spits out have a tiny Grok watermark in the corner. It’s a very small attempt towards labeling the image as AI that can easily be cropped out, but an attempt nonetheless.
Thankfully, Grok’s generated images still have a few telltale signs that something isn’t quite right. The image I asked for of Donald Trump and Kamala Harris on a debate stage gave the president-elect seven fingers while Harris only somewhat looked similar to the vice president.
Still, AI is becoming increasingly better at replicating real photographs. The image of a group of young adults on a city street I asked Grok to generate (above) gives them realistic hair texture complete with flyaways. The most noticeable sign that the image is faked is asymmetrical fashion with mismatched earrings and buttons that don’t match up with button holes.
In a digital world with a short attention span of constant scrolls, how many viewers will take the time to vet whether or not an image came from a camera or an AI? As AI not only gets better but more accessible with free tiers, viewers will need to be that much more diligent at spotting fakes.
You may also like…
Learn how to tell if an image was AI generated in our guide or read on how generative AI is becoming mainstream.
