Like many people interested in the advances in AI technology and concerned about whether or not robots are coming for their jobs, we were curious about testing out DALL-E 2, an AI-powered tool that generates sophisticated and detailed images.
With subreddit boards chock-full of creative and head-scratching image requests brought to life by the AI, DALL-E seems like a fun tool that might have a meaningful impact in the future. While we initially wanted to integrate DALL-E with News API to see what kind of images could be automated, we realized DALL-E doesn’t have an API yet, so we opted for another method to test out this AI’s artistic abilities.
By feeding DALL-E viral headlines of the most popular recent news stories we found by using our News API, we were able to compare the images the AI produces side-by-side with the actual images that accompanied the news pieces. Keep reading to learn more about what DALL-E does, how we expect it will be used in the future and our side-by-side comparison of the images. Just a warning though… once you see these images, you won’t be able to unsee them.
What exactly is DALL-E?
Named after the loveable Pixar character, Wall-E, and the surrealist painter, Salvador Dali, DALL-E is an artificial intelligence program that creates images from textual descriptions, launched by OpenAI in January 2021. Using NLP technology, DALL-E is able to generate images in different styles such as photographs, drawings, paintings, or even abstract art. When inputting text to create an image, you can be as specific as possible by describing the type of photo, its angle, what elements to include, how a subject should be posed, colors, artistic styles, and more. This technology is open to the public, and while this is only one example of technology that uses AI and NLP to generate content in development, its impressive capabilities make DALL-E a hyper-advanced AI tool that has the potential to be used in a number of different scenarios.
How will we use DALL-E in the future?
As a relatively new program, many people are using DALL-E now just to test the limits of the AI’s capabilities. With further development, there are several ways AI-driven tools like DALL-E can be used in both creative and non-creative businesses.
- Concept art – DALL-E is a cheap and quick way to create concept art like 3D models and illustrations for industries such as interior design, video games, and even storyboarding for movies or TV shows.
- Commercial images – Images for things like commercial children’s books, commercial newsletters, or email blasts can use DALL-E to create quick accompaniments to a written text.
- SEO-based content – It’s no secret that the Google algorithm loves unique, relevant images to accompany any piece of content. DALL-E can be used alongside your SEO-driven content to create royalty-free images that are specifically related to any topic.
- News item images – Dall-E can fill in the gaps for those news items that are difficult or impossible to have photographed due to being more abstract, such as images of space, hypothetical situations, graphics, or historical images that were never committed to film.
DALL-E meets News API
We took a simple approach to test out DALL-E’s capabilities. We used our News API to run a search of recent general news headlines. Our News API was able to rank them by virality by looking at metrics such as how many likes certain articles received. With a list of headlines ready to go, we selected a handful that was the most descriptive in order to input into DALL-E to see how the existing article’s main image compared to one generated by the AI.
We also took the liberty of rating each side-by-side image for the purpose of understanding what type of descriptive language works best when working with DALL-E and where things can be improved.
The results speak for themselves.
News Headline: “Shia LaBeouf converts to Catholicism after studying for ‘Padre Pio’ movie”
Rating: 6/10
It’s clear that DALL-E isn’t sure who Shia LaBeouf is. Still, this is a good example of how news items might use DALL-E to create visual content for articles where the topic is more conceptual.
News Headline: “Drought reveals new dinosaur tracks in Texas”
Rating: 8.5/10
An image like this is a great way to test DALL-E’s capabilities. We thought the DALL-E image was a little too cookie-cutter and a real dinosaur track might be rougher around the edges. That being said, there’s no reason an image like this couldn’t be used in a real article.
News Headline: “Snoop Dogg gifts Warriors’ Steph Curry iconic Death Row Records chain”
Rating: 8/10
This image is also a good example of how news agencies can use AI-generated images that are more specific to their articles. While DALL-E’s has been quite kind to Snoop Dog’s age, there is some resemblance, and the so-called “fake” image actually shows both Snoop and a chain, while the actual image doesn’t.
News Headline: “White mother with half-Jamaican partner has twins with completely different skin tones”
Rating: 3/10
DALL-E, we have questions. For starters, have you ever seen a baby before? The creep factor here outshines DALL-E’s capabilities. Obviously, when it comes to real-life portraits, there are improvements to be made. However, we did appreciate that DALL-E didn’t assume the couple was heterosexual, which further highlights how specific you need to be when telling DALL-E what you want in an image.
News Headline: “Blake Lively, Ryan Reynolds Family Photos”
Rating: 1/10
Ryan and Blake are one of those couples everyone loves to love, but the DALL-E-generated versions of them are straight out of a horror movie. We’re unsure what went wrong here, but it seems like the description was too short and we’re becoming convinced that DALL-E isn’t very familiar with celebrities.
News Headline: “San Diego Zoo welcomes birth of adorable white rhino calf”
Rating: 6/10
Like the dinosaur track, the rhino here looks a little too clean. It’s not an overall bad image, but the original image showing the rhino alongside its mother is more appropriate for the article.
News Headline: “Seal breaks into New Zealand home, traumatises cat and hangs out on couch”
Rating: 2/10
Aside from celebrities, DALL-E apparently also doesn’t know what seals look like. This image might have turned out better if the description was more specific and didn’t mention three different scenarios, though we would have loved to see the generated image of a seal traumatizing a house cat.
News Headline: “Mom posts photo of baby girl who looks like Woody Harrelson”
Rating: 1/10
DALL-E proved once again it has difficulties with celebrity faces. We’re unsure why DALL-E chose to turn a Woody Harrelson into a Benajmin Button-like situation where he’s suddenly a toddler with the face of an adult, but like a car crash, we’re finding it very difficult to look away.
News Headline: “8-year-old boy finds huge tooth from shark that lived 22 million years ago”
Rating: 6/10
The child’s face here is clearly distorted, though not as bad as other children’s faces we’ve seen DALL-E create. The level of detail though is impressive, and if the face was corrected it would be an entirely believable image.
News Headline: “The hilarious moment a young child is left totally starstruck by Bayern Munich’s Sadio Mane”
Rating: 8/10
Finally, DALL-E understood what a child should look like! There are some aspects of this image that make it questionable, like the cleanliness of it along with the strange pixelation of the child’s hair, but overall, it’s a pretty spot-on depiction of a child’s excitement in front of their idol.
News Headline: “Cambridgeshire man, 92, passes GCSE maths exam with top mark”
Rating: 8/10
We wanted to rate this one higher since it’s not a bad representation of the title, but the title stated the exam the man passed was for math, and somehow DALL-E generated an image of a man sitting behind what we can only speculate is an Elvish language exam.
Is DALL-E coming for your job?
Graphic designers, visual artists, and photographers can breathe a sigh of relief in knowing that DALL-E isn’t going to replace them anytime soon. While the tool is highly impressive, it won’t replace human creativity or professionalism.
Still, DALL-E demonstrates the impressive capabilities of using AI and NLP technology to create something out of nothing. Certain aspects of Webz.io’s News API also rely on NLP technology to gather sentiments on a huge variety of news articles. However, for tools like News API, that technology is used to help companies gather intel on what is said about them online for the purpose of analysis and reputation management. With DALL-E, the technology is creating nightmare-inducing photos of babies, at least for the time being.
