AI has busted into the art scene in a big way recently. Companies like Nvidia have been working on AI tools to make art out of basic squiggles (opens in new tab) while robot AI artists do unique and interesting paintings. (opens in new tab) Lately, the internet has been abuzz with images created by the DALL-E 2 art bot that can create images from just a few words, likewise the more lightweight DALL-E mini homage. They are often a good mix of impressive and utterly cursed, much like these AI-generated Duke Nukems (opens in new tab).
While amazing for generating fresh mematic content on the regular, these AI art bots are at their hearts, tools, and one photographer (spotted by PetaPixel) has found a great use for the software. Rather than creating a whole image from a description in DALL-E 2, they’ve used it to clean up a photo taken, and the results look more impressive than Photoshop.
Nicholas Sherlock (opens in new tab) is a photographer based in New Zealand who used DALL-E 2 to turn some very nice but blurry photos of a ladybird, into crisp stock photography style images. He did this by having the AI edit his original blurred photo with the text prompt “Ladybug on a leaf, focus stacked high-resolution macro photograph.”
Sherlock then showed the images to photography YouTuber Micael Widell (opens in new tab) who featured them in a video exploring DALLE-2’s use in and potential murder of photography. This video goes through a few different examples of DALLE-2 before getting to Sherlock’s example in photo editing, and all the results are at least somewhat terrifyingly impressive.
The video also showcases another example where DALL-E 2 was asked to generate other variations of a provided image and the results seem impossible to discern. The capabilities of these AIs for image generation is staggeringly impressive. PetaPixel also compares more images provided by Sherlock for comparison. When shown against images that had fixes generated by Photoshop, DALL-E 2 came out the clear winner.
The ability to flawlessly edit an actual photo in a desired way does seem like a great use for the tool, especially with results like this. It’s fair that Widell wonders in his video what the role of a photographer or even hobbyist will be in a DALL-E 2 AI future where anyone can generate realistic images of whatever they want.
Of course there will always be other reasons to pursue these passions, but when it comes to things like stock photography especially, these AI’s certainly feel like game changers.
It’s likely we’ll start seeing an increase of AI generated art being used in games too, be it just to get concepts out and explore options, or even making usable assets. There are plenty of questions about what this means for the future of visual art in many industries and hobbies.
However, it’s important to never underestimate the human desire and joy in creation. Hopefully these AIs can be a tool to empower that more often than they take it away.