Yes. Are human artists redundant? No.
I chatted briefly about text-to-image AI programs on my podcast, but have had a few more thoughts.
One of the reasons that I was promoted early in my career was my ability to create macros. Unlike their name, macros are small routines that allow a computer to do repetitive tasks. Mine were created in a long forgotten spreadsheet software called Lotus 123.
Software like Adobe Photoshop also allow repetitive tasks to be automated. It’s currently called ‘Actions’. Actions allow a filter to be added to a photo, so you can import a photograph and turn it into a charcoal drawing.
Now, in 2022, the software has become smarter. It’s been trained to be able to translate text into images. Just as speech to text software has become better through millions of interactions e.g. Siri or Alexa, so too is text to image software becoming smarter and smarter with every new image created.
The most basic of these programs effectively do a Google image search and combine the results into a single picture. Just like Google, the more specific your search term, the higher the chance you will find something relevant. In SEO, they call that the ‘long tail’. A search for ‘man’ will probably not get you the results you want. A search for ’49 year old Australian man at the beach’ may return a better result – if… someone has tagged that image as such.
The current text-to-image AIs work in the same way. Type ‘Man’ into an AI and…
Not exactly Michelangelo’s David, and the piece of tape or sticking plaster is a weird addition – perhaps there is something the AI knows that I don’t. Otherwise, it is what it is.. Man. Note the inherent racial bias. It’s a white man.
So it’s a case of ‘Garbage in, Garbage out’. If you can’t describe what you want in detail, you may not get what you are looking for. I’ve experimented with creating a self-portrait, a bit like a police witness statement. Some of the current AIs are very bad at faces and many are not iterative.
The Midjourney AI, which, based on my use, is currently the ‘best’, produces some stunning images, but it’s never seen a picture of me, so it can only go with what I give it as a description.
The 4 images on the right are the first go. Not me, but definitely what was described – perhaps a little too old? Midjourney allows you to pick one of the first thoughts and ‘expand’ on it. Then you can ‘upscale’ it, which means add more detail.
It’s a 49 year old surfer with green eyes and reading glasses. It’s not me, but I haven’t surfed for a long time. So my neck is not that thick!
Not sure what this guy did to look so glum. This is close-ish.. the kind of thing you might get if you paid an artist on the street $20 do do a portrait that you sit for 10 minutes for.
It’s an extreme user case. And the results are very good considering how young these algorithms are.
But can these AI’s create art? Yes…
I asked the same Midjourney AI to ‘imagine’ a Yacht Race in Newport Rhode Island on a stormy day as an oil on canvas and this is what I got…
I love it. I’d pay for it (If it was indeed oil on canvas). Sure, it’s derivative. But it has to be. There is no way an AI can create an image without being trained by other images. So what would yacht racing on a sunny day in Sydney as a pen and wash image look like?
Are human artists redundant? Perhaps the bad ones.
By definition, an AI is derivative. Many examples of the output on social media actually specify a painter’s style e.g. Picaso or Van Gough. Some of the more basic ones, like Dream by Wombo.ai actually have these ‘filters’ or styles built in, like Instagram. You can choose from Comic or Surreal or Baroque or Steampunk or Psychedelic.
There is a new skill or talent though. To get the best results, you need to be able to type what you want, in English, in detail. You need to know the difference between oil on canvas and watercolour. You need to know that you can specify a 16:9 ratio or what the light source is. This is a tool. And experts will be able to use the tool better than non-experts, the same way a professional photographer takes better pictures with an iPhone.
There are some interesting implications for copyright and other intellectual property rights, but there are also some great applications, especially in marketing, like the creation of mood boards and visualising a brand.
It’s just getting interesting…
No Comments