AI Art Generators, Explained

By | October 21, 2022

Artificial intelligence has made it nearly effortless to create digital art.


Foreign [Music] This is your Tech news briefing for Friday October 21st I’m Julie Chang for The Wall Street Journal filling in for Zoe Thomas Chill intelligence has made it Incredibly easy to create original works Of digital art programs like open ai’s Dolly 2 and stability ai’s dream Studio Let you type in almost any phrase to get A computer generated image and while This can be a useful and even fun tool Some worry about the potential drawbacks Or even dangers our senior personal Tech Columnist Joanna Stern has tried out Both programs and she’s been looking Into what they could mean for artists And anyone coming across those images Online and she joins me now hi Joanna Thank you so much for being here anytime Joanna you tried out two of these AI art Generators how did those images turn out It really depends I will tell you that You will always get something back That’s a little bit entertaining and Probably humorous and sometimes you’ll Get back stuff that you just can’t Believe was made by artificial Intelligence and not a human I’ll talk About a couple examples that I used in The column one was a technology Columnist who was trying to write a Column those were the words I put into

Dolly and what I got back was a picture Of a man his face was a little bit Deformed and he’s writing with a pen on A keyboard he’s also wearing a very ugly Purple tie Another one I tried was a female Technology columnist in a spacesuit Typing on a laptop in outer space And when I got back was a very cool Image of outer space sort of glowing Stars blue background behind this person And then the person in an astronaut suit Typing on a laptop and that image is Very cool I would probably make it my Phone background yeah that image does Sound really neat but how exactly does This technology work and where do they Get the images so it’s very complicated Technology and I spoke to many experts To just try to make this as simple as Possible for myself and for my readers And my viewers and what I was really Taught and what I think is the best Comparison is that the artificial Intelligence which is living on the Servers of these companies Is being shown flash cards The artificial intelligence is learning By looking at millions and millions of Flash cards and what’s on those Flashcards well it’s images and then It’s labels so it’s a picture of Something in let’s say a monkey and it Looks at thousands of pictures of

Monkeys that are labeled monkey and so The AI starts to learn that is what a Monkey is and then it looks at some Pictures of podcasting right maybe That’s a microphone or people talking Into microphones and then through all of This deep learning and all of this Technology when you put in the phrase a Monkey recording a podcast the Artificial intelligence starts to build That image pixel by pixel and it’s not Putting together images it’s already Looked at it’s building that image from Scratch are there any limitations to it Though for instance is there anything That they wouldn’t let you create so yes One of the images that I love that I’ve Made is of Elon Musk holding a Twitter Bird as you would expect you Elon Musk Would now have as a pet but I wasn’t Able to make that image through Dolly Because Dolly restricts making images With people who are public figures they Have some other restrictions too which We can talk about I was able to make That image though using stability ai’s Dream Studio that image generator Doesn’t have restrictions on using Public figures and so when I put that Prompt in I got back quite a realistic Looking version of Elon Musk it doesn’t Look like a photo of him it’s very clear It’s not really him but there is enough Similarity there that when I’d tweet

That photo I don’t need to say who that Is people would know that’s an Elon Musk Look alike and what are some of the Other restrictions with Dolly yeah so Both Dolly and dream Studio do have Restrictions on making explicit content Right you can’t really make porn with These they also have restrictions around Showing really offensive or quite vile Lint content Dali specifically limits The ability to generate violent hateful Or adult images I did try on both of These systems The Prompt a photograph of A terrorist attack and both of them did Produce images though the dolly one was Definitely less offensive or less in Your face on some of the imagery it was Of a police car it was a little bit Stylized there was some sort of fire in One of the backgrounds but on dream Studio it definitely went a step further And it showed deformed bodies on the Ground with guns and fire so there’s a Difference between the two systems but There’s also this bigger question about Which of these companies will come along And not put these types of restrictions In place Were there any signs that maybe these Images were AI generated yeah the Biggest indicator that something is Generated by AI right now is the quality If you look closely at pretty much Anything you can see that there are some

Issues some warping of the image Especially if it’s trying to mimic a Photograph or it’s trying to look more Lifelike and so with those terrorist Images the heads weren’t really quite on The bodies correctly there were Definitely guns but it didn’t look like They were being held in the hands so There were a lot of signs in that image That this wasn’t real but everyone I Spoke to for the story said this is Going to get so much better so fast and So that’s the big question going ahead As we start to get images that look like They might have been taken by a Photographer how do we know the Difference openai is really encouraging People to say and credit when they share Their images that it was made by Dolly They also put this watermark in the Bottom right of the image so you can More easily know that this was made by Dolly but of course you can crop that Out stability AI does not add a Watermark I want to talk about some of the other Potential downsides to this technology Specifically about how this may impact Artists how have they responded to AI Art taking off So I’ve heard mixed reactions from Actual creators obviously this Tech is Good right now but it’s going to get Great and so I think the bigger

Questions about the future and how People use this but right now some Professional illustrators and animators I’ve spoken to are actually using these Tools right they like the idea that they Can make certain things by typing in Prompts and then integrate it into their Works if you’re an artist and it’s Ripping off your style well then you’re Not going to be happy you know another Potential issue is bias Tech has been Criticized for having bias in it did That show up at all while you were Testing out the program yeah it Definitely did and I’ll come back to one Of the prompts that I put in in the Column which was a technology columnist Writing a column and all the images Returned in that prompt were of white Men sitting at computers or writing on Notebooks and as a technology columnist That is not a white man I felt like hmm Well I guess I’m not really represented Here so the companies told me that this Is something they’re working on Specifically open AI which owns Dolly Said that they continue to invest into Research on mitigating bias and improve Results all right Joanna we’re gonna Have to leave it there our senior Personal Tech columnist Joanna Stern Thank you so much for breaking this down For us thank you and that’s it for Tech News briefing this week our supervising

Producer is Chris sinsley and I’m Julie Chang filling in for our host Zoe Thomas Thanks for listening and have a great Weekend [Music] Foreign

Leave a Reply