Pay what you think it's worth.

A year long social pricing study to understand the value people place on photography.

AI-generated art photography is here, but it’s not going to replace your camera


It seems like AI-generated art has suddenly exploded everywhere in the last month. A friend showed me what he was creating with DALL·E 2, and down a rabbit hole I fell. And man, what a fun rabbit hole it is too for photography and illustration alike!

AI-generated art is literally that: Artworks created by an AI generator. Fed millions upon millions of images and trained by humans and machines, these tools can produce incredibly detailed, sophisticated and realistic images based on natural language text input. As in, I can type a sentence like “photograph of a mountain lake scene” and get an image like this.

I generated this image using Midjourney, an AI engine accessed in Discord.

There are parameters to learn and different algorithms to master, but it’s a pretty shallow learning curve before you can produce some cool stuff.

Midjourney leans toward an illustration style of image (at least for me — it learns from your generation process) but you can get it to produce images that are photo-adjacent, if not photo-realistic.

Computer-generated art but on a whole new scale (that’s actually good)

I’ve been playing with computer-generated art since 2001 when I installed a fractal generator on my family’s first PC (it had a whole 4GB of space on the hard drive!) and used it to write an essay for my advanced math class in school. I’ve generated plenty of those wispy-color-on-black fractal flame images that are so 2000s, reminiscent of the visualizer in Winamp (remember Winamp?!).

This one I titled “Blue and Pink Diagonal Swirly” (2005) which I think we can all agree is an apt name. It was created with Apophysis, which, I was surprised to find, still exists.

But AI-generated art is a whole other ball game. The things I create actually look like the things I want to create. It’s a little mind-blowing when you get your first few great images.

There’s something deliciously uncanny-valley about the images. The AI interprets what you’ve written in sometimes bizarre and strange ways, making pictures that send your head tilting to the side as you say, “Well, it’s not wrong, but it’s not how I pictured that …”

I wanted a macro photo of a bee, and I mean, it’s not a macro photo of a bee, but … it’s not not a macro photo of a bee either.

Some things are easier for it to understand than others. The misty forest below is pretty solidly landing in misty forest town, but my husband’s (huge) list of demands of things I had to create included a lama made of pasta, a walrus dressed as a sausage, a school bus full of nuns, a depressing yard sale, a mechanical bee, a haunted ant and the contents of a void. These were met with mixed success (I’ve shared some on my Instagram if you really, really need these images in your life).

The photograph-style images produced by AI generators are more reliable when you write a landscape prompt.

AI-generated art platforms for photographers

The number of platforms entering this space are growing. What they all have in common is cost. It takes a lot of computing and this doesn’t come cheap. Here’s a quick overview of the platforms I’ve found so far.

DALL·E 2 AI generator for photo-realistic art

Perhaps the most talked-about for photo-realism is DALL·E 2. In mid-July 2022 they opened in beta and are onboarding 1 million users from their (long) waitlist. (I’m still on the waitlist.) I’ve heard of people waiting months and others getting invites within days.

Their onboarding process is not first-in-best-dressed and some categories of users get preference over others. DALL·E 2 gives you 50 free credits your first month, then 15 every month after that. Beyond that, you need to buy credit packs to top you up.

I’m excited to give DALL·E 2 a try and see how it can be used to create AI-generated art photography images. Stay tuned!

NightCafe AI generator with lots of different algorithms

NightCafe is another one that gives out free daily credits. I have had limited success getting images I love from it. I am probably not throwing enough credits at my images. It gives you 5 credits a day, and a bonus 20 for completing your profile. They have just added Stable Diffusion to their algorithms, so I’d expect the possibilities to grow here with time.

Stable Diffusion’s photo-realism is a little more photo-realistic than Midjourney.

Stable Diffusion, newly released on the AI generator scene

Speaking of which, I’ve just started trying the Stable Diffusion algorithm on its native platform. It has a cap on a resolution of 1024px but pleasantly shows how many of your credits you are using to generate each image. You’ll get 200 credits to play with here before having to pay (each of the images above took about 15 credits).

This is another Stable Diffusion one — the night sky is stellar, compared to Midjourney.

Midjourney, illustration style AI-generated art

Midjourney is my obsession at the moment. As a fiction author, the chance to get words from my head into real illustrations is irresistible. It’s accessed through Discord so you have to have some familiarity with the platform to use it. You’ll get roughly 20 free image generations before having to sign up for a plan.

(Side note: If you think you’ll want to have a decent play at Midjourney, skip the entry-level plan and grab a month of their unlimited plan. I used all the processing units on the basic plan in three days when I first signed up.)

Midjourney has a great illustrative, real-but-a-little-surreal look which I love for my AI-generated art.

Who owns the AI-generated art and photography you create?

Concerns have been raised about the ethics of AI-generated art. With algorithms trained on millions of images, it’s not clear whether or how original creators can be credited. These discussions have been going on for some years and they will not be resolved soon.

From a technical standpoint, AI generators are considered “tools” in the same way that a camera is a tool. The person using the tool (i.e. me or you) is the creator and copyright defaults to the creator. That is unless otherwise agreed (e.g. by terms of use tick box when you sign up for a service!).

It’s pretty hard to look at an image like this, created by an algorithm, and not wonder where it came from.

Midjourney, and now DALL·E 2 too, give you the creator commercial rights to the images created. On DALL·E 2 this is a recent change to their policy. This suggests it’s likely to become the standard across the board as AI-generated art becomes more widely accessed.

What does AI-generated art mean for photography?

As you try out different platforms you’ll see that some are better than others at photographs. Sometimes Midjourney interprets my “photograph” or “photo” requests literally. It returns something best described as a scanned photograph from the ye olden days, complete with creases and dust scratches.

A literally interpreted photo generated by AI, complete with dust marks, creases and discolored paper.

As technology develops, the accuracy of real-world things will improve. Things a little “That’s not quite right,” now, may become indistinguishable from real photos.

It’s not a moth, but it’s sort of a moth.

As I play with image generation I keep thinking about the invention of the camera. How did portrait artists, who studied at the feet of masters for years and years to perfect the art of painting pictures of people, feel about the invention of the camera? The camera effectively turned a long and arduous process into something quick — eventually instant. That feels like what’s happening here.

Is AI-generated art photography the end of photos?

Luckily, there are huge limits to the technology that means AI-generated art will never replace photography entirely. Aside from the filters that exist on all the platforms to prevent the generation of harmful and offensive content (“Make your mother proud,” says Stable Diffusion in their onboarding pop-up), you can’t create real life people, or reuse elements in repeated generations.

For example, it’s not possible to generate a character like a green furry robot (let’s call him Jim) and then reuse Jim the Robot in a bunch of wild and wacky images depicting his adventures.

This AI-generated macro photograph of a beetle looks cool … but it’s not a real beetle.

And it’s, obviously, just not real. Photography is all about capturing real people, real emotions, real landscapes, real moths, real bugs and real llamas made of pasta (well, maybe not that). That will never be replaced by AI. So go play, explore a new medium, get ideas and go off on tangents. Your camera will be waiting when you get back.





Source link

Rafael Jones

Back to top