• Images vary from the presentable to the horrendous, leaving room for creatives
Who would have thought there would come a time when, by punching a few words into some website, one could generate very realistic images of just about anything?
That beautiful woman you see online, that enticing holiday destination and that adorable home may have been created in a computer's electronic circuitry. The software can create pictures of things that don't exist in real life. What's more, some websites can generate such images for free, making the technology available to everyone.
All you need do is tell the website what you want the picture to look like. It's a bit like having an artist before you listen to instructions on what to draw. Intelligent software has been taught to interpret human instructions and make pictures that look real. You can add effects, too, whether it's strong sunlight, cloudy conditions, rain, snow and whatever colour you prefer.
The implications of these developments raise a critical question. If anybody can produce pictures from out of the blue, how do we tell apart the truth from lies?
A good example is the picture of a Toyota Land Cruiser on this page. The picture was generated from a text-to-image website using the keywords: "Grey Toyota Land Cruiser on a wet city street with evening sunlight". Within a minute, the website generated several images for one to choose from.
The next step was to find out if artificial intelligence merely copied an existing image. A reverse image search on Google and Bing, two of the most popular search engines, did not find a similar image anywhere on the Internet. This little experiment proved that artificial intelligence generated something that has never before existed.
The next trial was to generate an image containing human beings. The keywords fed into artificial intelligence were: "African men and women cultivating a maize farm with hills in the background in sharp sunlight". The result was a fairly presentable image, though the human forms have a somewhat strange appearance. The maize field and hills in the background look real. A reverse image search did not find similar scenes anywhere else on the Internet.
Will artificial intelligence put photographers, graphic designers, illustrators and similar artists out of work? These are people who make a living by producing creative designs.
Brian Leonard is a Nairobi-based graphic designer who does not see artificial intelligence as a threat to his livelihood. "Artificial intelligence has made my work easier and it's raised some standards," Leonard says. Though graphic designers have been using computers for many years, Leonard agrees the profession is experiencing massive changes.
"When I was starting out designing, we were heavily reliant on apps such as Photoshop to remove backgrounds and add lighting. Now there are websites that do that within seconds," Leonard says. He can now work on images faster and move on to other tasks.
Another example of artificial intelligence making work easier is in the making of animated images, such as cartoons. In the past, cartoon characters used to be drawn by hand and then scanned into a computer to produce motion. The process is now faster and easier.
"Artificial intelligence has developed to the point you only need to feed which details you want out of a cartoon character. I can key in a character with a red shirt and brown shorts, describe the facial features I want and the software translates that into pictures," Leonard explains.
AI has raised standards in the profession, which may be contributing to some people feeling threatened by technology. "Are you willing to adapt and use it to do more, or simply be the one to complain that it has taken away your role?" he asks.
Are you willing to adapt and use it to do more, or simply be the one to complain that it has taken away your role?Brian Leonard
There will always be a need for professional graphic designers because artificial intelligence has weaknesses. Some of the results that emerged while testing image generation platforms were horrendous (see examples). From one website, the keywords "African man with Japanese woman" generated a couple with mashed-up fingers. One picture had a woman with three legs!
Here's a hint of what to look for if you suspect a picture was generated by artificial intelligence: check out the hands. Artificial intelligence has difficulties drawing human hands. It is a problem acknowledged by experts in the field. It all goes back to how the software was "taught".
Dr Jim Fan, a scientist in artificial intelligence, says there were few detailed pictures of human hands from which the software could learn. "Face pictures are much more common than hand pictures," Fan says. "Even when the whole body is shown, hands tend to occupy a much smaller space." Furthermore, hands are frequently hidden by other objects.
Producing African faces is another weakness. Various text-to-image websites were asked to produce a picture of African men and women working together on a farm. Of the websites tested, only one produced a decent image. The rest generated humans with distorted faces, missing limbs, extra limbs and ghost-like bodies. This problem may arise because the software does not have enough information on dark-skinned Africans compared to people of other races.
Those weaknesses are a big reason designers cannot completely rely on machines. As the Kenyan designer Leonard tells us, machines need guidance. "There's no need for you to draw by hand, but there are parts which a machine will definitely miss because nothing is ever 100 per cent," he says. In his view, graphic designers should aim to do things that computers cannot do, such as making unique paintings. Collectors are willing to pay for good-quality art.
With artificial intelligence easily within reach, how can we be sure of anything we see online? How can you be sure the car, house or land advertised for sale actually exists? How can you be sure an event portrayed in the media happened as reported? It takes lots of scrutiny to detect a fake image. Ultimately, where money is involved, it is better to visit the property and confirm its existence.
Want to try making pictures from artificial intelligence? Some of the commonly used text-to-image websites include MidJourney, Picsart, OpenAI, Dreamstudio and Nightcafe. A few are free but most require a paid subscription to fully utilise the features.