DALL E 2 is a new artificial intelligence system that can create realistic images and art based on a written description. No more complicated styling and lighting setups: you can now just describe what you want, and DALL-E-2 delivers the picture. Too good to be true? Too threatening to the ever successful photography industry? Convince yourself with my prerelease test run.
DALL-E-2 states in its mission:
OpenAI’s mission is to ensure that artificial general intelligence (AGI) – by which we mean highly autonomous systems that outperform humans at the most economically valuable work – benefits all of humanity.
If you’re a photographer who feels that modern technology is constantly scrutinizing your ability to be a vital player in the marketing industry, then this statement is definitely one to let you down. I got pre-access to the artificial general intelligence (AGI) platform and took it for a test run. Can it really do what we can? Can it even “exceed” us? Is it a threat to the photographer? Is it a tool? Or is it a combination of both? Let’s see.
There are a few features of the software. The first, and what it is most famous for, is that it can generate an image or artwork based on a description. For example, on their Instagram you will find the result of “a blue orange cut in half on a blue floor in front of a blue wall”
Everyone can agree that the result is quite stunning. I even attempted a random description myself.
There’s no denying that the technology is impressive. However, my intention in pre-testing was to find out if a professional photographer could do it function. Instead of hiring us, could a client type the description of what they wanted and skip the cost of hiring a professional?
Test one: are the images generated comparable to those of a professional photographer’s work?
My first test was to see if DALL-E 2 could generate visual content that could compete with the images I was working on at the time. Case study one: a chocolate made with cocoa and dates. I typed in the description of the image I took that morning: “a date with chocolate sauce poured on it.”
These were the results:
I suppose if you just need a picture of dates with chocolate, this might suffice. However, if you were to consider lighting, composition, color correction, or aesthetics, these images would not meet my standards.
Then I decided to throw a model into the test. The brand had once taken a picture of a model dripping chocolate on her tongue, and it was a very successful image. Along those lines, I typed, “A beautiful woman with chocolate drizzle all over her body.”
My first observation was that it seemed like artificial intelligence picked white brunettes as their portrayal of quintessential beauty, so I guess I’m out of luck! My second observation, as in the previous test, was that the aesthetics of the images were a complete fail. It looked more like a scene from a Freddy movie than an ad selling chocolate and lust. The software impressed me in that it could magically generate images from a short description, but it soon became apparent that it was in no way capable of creating a cohesive set of aesthetically successful images.
Test Two: Can the Correction Features Benefit the Photographer?
You may have seen DALL-E 2’s almost improbable results from the AI-corrected blurry ladybug, as seen in this Tech Times article. I decided to give these features a try as well. My first trial was to remove a shadow and fill it with a patterned background. I think I jumped right in at the deep end.
After uploading my image, I selected “Edit image” and typed “Remove the shadow from the skincare bottle and fill it with the palm leaf shadow”. I was undoubtedly impressed by the images it produced.
It significantly outperformed Photoshop, which couldn’t match the palm pattern.
For the amount of criticism I’ve dished out thus far, I really have to take my hat off to the software on it. Then I tried another realistic scenario. Once, my salsa client asked me to swap the red peppers in the image below for jalapeño peppers. Needless to say I had to shoot it again. I was impressed with DALLE-2’s latest correction and decided to see if it could get the job done.
“Change the red peppers for jalapeño peppers.”
“T, To May!”?… and the peppers are still red.
A clear failure on this assignment.
Test Three: Can Dall-E-2 Effectively Add Elements to a Photographer’s Image?
In my product photography I often make a lot of splashes and crashes. My final test was to see if the software could do some of that work for me. Inspired by the images I took below, I asked if it could add chips to a background.
Here is the result for “Add tortilla chips to the background.”
I also asked the software to add more water drops to a shot.
Below is the result for “Add a splash of juice to the background.”
The test above did not generate a splash and some interesting alternatives, such as a hazy pineapple creeping in.
After DALL-E-2 had to endure a myriad of challenges, it was clear that the software hadn’t yet fulfilled its mission to “exceed” a professional photographer. While the software is an incredible feat, it doesn’t always deliver what it asks for. If so, the aesthetics of the image are not up to par. I was amazed at the palm shadow repair and wonder if it will position itself as a more advanced tool than Photoshop.
What do you think of this new technology that aims to “exceed people at the most economic value”? Share your thoughts below.