2 common mistakes people make when generating AI images, according to Picsart executive

• Gen AI images have become increasingly popular in the 2020s.
• Tools have become more sophisticated since first launching a few years ago.
• But people are still making common mistakes when generating photos.

Rewind a couple of years and you probably never would have imagined that you could create a photo on your phone or laptop just by typing out a few simple words. But by the time 2024 has rolled around, we’ve probably all dabbled with generative AI (artificial intelligence) at least once or twice.

If you’ve tried any of the number of advanced tools to make your images and have been left disappointed by the results, you might be making simple mistakes that are holding you back. To find out how to improve your results when using these AI-tools, I spoke to the co-founder and COO of Picsart, Mikayel Vardanyan, at the World Congress on Innovation and Technology in Yerevan, Armenia, this weekend.

He also offered his verdict on what might be next for generative AI - and what companies can do to stand out in the crowded market.

The simple mistakes people use when making AI images

During an interview at the WCIT in Armenia over the weekend, Picsart’s COO told me there are two simple mistakes people usually make when trying to generate AI images. And it is probably not what you are thinking.

Mikayel explained: “It's not so easy to prompt to get the result that you expected. To say, I think it's first of all a matter of expectations.

“You expect whatever we write, they're able to just take and generate super nice stuff. You'll be happy and just take it and do it. That's one thing.

“And second, it still requires a lot of effort to come up with the right words, with the right stuff, to generate something plus on top of it, not that you just generate and you are done. Usually you also need to do post editing and do something yourself without prompting.

“And I think that these are the common mistakes, expectations and how to prompt correct and do not rely on it completely, yeah.”

How to improve the AI images you are coming up with?

So, we’ve identified some of the simple mistakes that you might be making when trying to generate AI images, but what can you do to actually improve your creations. And it is not just taking extra care to craft your prompt - although that will also help.

Mikayel said: “So easiest will be to experiment with filters, experiment with some retouch tools, experiment with removing backgrounds and generate backgrounds. So that's the easiest.

“If you do not have any ideas in mind, you still can go there and be inspired by the list of community created photos. And in certain cases, you can see how it was created and learn.

“That's what we call replaying, learn what was done, which tools have been used, and replicate the same, but with your photos or with your ideas.”

But what about if you are taking things a bit more seriously, perhaps generating an image for business purposes. He continued: “It depends on your needs. So if you need to generate something and you need to replace something within the photo with your face or your object you are selling, that says one thing.

“Then, yeah, you generated something. Then you need to polish it or refine it. Then it's filters, then you want to print it, and you need quality generated stuff.

“You need to upscale it. So that's why, depending on the needs, we provide all the toolsets for you to understand what to use post editing if you, of course, understand and try to make it as accessible as possible for non pros, yeah.”

What will the future of AI generated photos look like?

In the couple of years since AI image generators first hit the market they have already come a long way - albeit they still may have a distance to go. But if you are reading this, you probably remember when people in the photos would have the wrong number of fingers and other obvious flaws.

Picsart’s COO believes that the next stage of evolution for these tools will come in the form of generative videos. He explained: “Think it will get more mature and mature, especially the video side of Gen AI.

“That's one thing. I believe everyone will work on optimising the performance costs. But another thing which will get better and better in terms of models, we already think it's commoditised. So everyone has their model, and there is no difference in many cases.

“And there are too many models even right now. So that is not a differentiator anymore in the form we currently have it, but the advising, the focus will be more on video, how to make it better, and also cost performance part.”

I attended the World Congress on Innovation and Technology from 4 October to 7 October. You can find all my stories from the event here. You can get in touch with me by emailing: [email protected]