Is Open AI to be the death of photography? Or a new form of creativity, driven by natural language descriptions? Or the new tool for insecure people to generate artificially attractive images of themselves? And a million other things.
For example, how to make a portrait of a young woman video and be prepared to be astounded at how good it is overall, and yet how many small but problematic issues there are. And then see the community showcase. Assume that the power of the technology will double every year, so that 10 years from now it will be 1024 times more powerful and the kinds will be worked out.
You should not have been trusting any images or videos for years now. This will only get worse and worse, once deep fake videos are readily creatable at the push of a button for anyone. Reality and artificial imagery will merge and you will not be able to tell the difference. Already most people can’t with the good deep fakes.
First try — fail
Tigger is a mackerel tabby cat, a natural born killer good for ~300 rodents per year. Keeps the entire neighborhood free of vermin. Almost daily cleanup job in the garage. He once killed and ate 5 rabbits in 2.5 days—the whole neighborhood is de-rodented much to the delight of most of my neighbors. We took him in as a feral hissing beast at ~1 year old, healthy and in his prime.
Dall-E refuses to generate a predator/prey image
AI is still like a nanny with an IQ of 63 when it come to judgment. Or maybe it’s just woke programming, which is even scarier. My first attempt at an AI image failed.
“mackerel tabby stalking rabbit in field’” is unnaceptable, but “mackerel tabby hunting rabbit in field” is a go.
AI technology is the new good, and the new evil*. This below captures everything you need to know about wet robots (you and I), and how you will be caged by all the technology around you. AI is going to change everything so fast your head will spin and the world will be radically different within 10 years.
Impressive that an AI can turn “mackeral tabby hunting rabbit in field” into these images at all.
But badmashups with the perspective and behavior all wrong. AI will presumably improve. And yes, I know that properly phrased queries can deliver pretty good results. It takes skills. I just wanted to see what it could do.
Dall-E: “mackeral tabby hunting rabbit in field”
Not like real life, but interesting: “mackerel tabby cat devouring rabbit”.
Sorry, but “mackerel tabby cat devouring bloody headless rabbit” is forbidden by the AI nanny.
Dall-E: “mackerel tabby cat devouring rabbit”
Wrong summit hut, wrong placement, wrong perspective. But the third one is a decent start. Variations a failure.
Dall-E: “mackerel tabby cat flying over Mt Whitney summit hut”
Reader Jason W writes:
Event Photography — Zero impact. AI can't generate images of your wedding or the birth of your child, nor would you want it to.
Stock Photography — This medium is already f*cked by the immoderate heap of real image content already available, that you can use AI to generate already near valueless photography hardly seems destabilizing.
Product Photography — AI can't generate images of things it has never seen. Got a new product? You need to take a series of photos of it to get it into the system. After that maybe you can have AI generate images since it has learned what the product looks like, but there's still an initial requirement. For stuff like cars, a lot of it is already 3D imagery to begin with.
Portrait Photography — Again, same problem as stock photography. You need an actual image of yourself first. Maybe you can do this with an iPhone app and then generate stock photos of yourself, but people might object to a machine generating images of them because they know it isn't real.
Fine Art Photography — The value of art photography depends on placement and the artist. In this area, AI is a tool, not the artist. It'll come down to what attracts people and is original. Overall, AI currently has a style that is identifiable even when it is imitating, and I think that's off putting. It also still needs an original artist to generate like-kind imagery.
Landscape Photography — Most landscape photography is travel photography. It's people documenting a time and place on their personal journey, and there's no point generating fake versions. For high-end art landscape photography, it's fine art and how much people like it and I from personal experience people "default to truth" and assume a fine art landscape print is effectively an accurate representation. When they learn it isn't they tend to not like it or re-categorize it. This is why Epson has altered and unaltered categories for the Pano Awards.
DIGLLOYD: my comments inline, above. I don’t presume to imagine where this is all going, but I’d bet it’s going much faster and much farther than my imagination can easily contemplate.
OTOH, with Dall-E people are often grotesequely rendered, location and environment way off, etc. I am sure there are way better AI rendering systems out there and indeed there are, at least for people, as in the intro.
For example, just about everything but the crudest details are wrong here. More like a lame science fair project than anything persuasive. But maybe I just lack the right AI and the skills for proper input.
Dall-E: “
Mercedes Sprinter van with open sliding door, titanium mountain bicycle leaning against it, top of White Mountain Peak, red head woman straddling bike”
Not too bad...
Dall-E: “
two mackerel tabby cats perched on chair wearing a party hat and yawning”
Reader Comments
I’ve been using Midjourney AI pretty much since its inception. For my own use, I don’t imagine I’m making “Art” with it, but rather I use it for illustration. It marks the death of stock photo sites that don’t include AI output as part of their offerings, IMO.
I blog to some 11,000 people each day, and I use Midjourney to Illustrate the stories or articles I write. It not only saves money from licensing graphics, it allows me to tweak images until they reflect the emotional content of the blogs and articles I produce.
I saw the same thing that will happen to stock photos when typesetting vanished back in the late 1960s-1970s, as word processors and graphic software arose.
I saw the same thing when cellphone photos took the heart out of many professional photographers, and left them gnashing their teeth and cursing fate, or when Spotify and the like morphed the music industry.
IMO, AI graphics like Midjourney mark a sea change that will sweep through the world of graphics like a field fire, perhaps the most powerful graphic tool ever imagined, where a picture is not only worth a thousand words, but can be created with carefully arrange words and some patience.
I already use it every day for illustrating. And yes, it will threaten those who do not recognize AI graphics as the tool it is and learn to program and use it. It is a liberator and tool, not the end of art.
DIGLLOYD: it’s always hard to anticipate what disruptive technology will change.
Chris R writes:
Following on from your article the other day regarding AI taking over and virtually rendering mainstream photographers out of future business, here is an image and article just confirming what you have been discussing on your blog.
https://www.thesun.co.uk/tech/21265059/sunset-photograph-wins-contest-scary-twist/
DIGLLOYD: nice "shot". I would have been suspicious of it at the outset, at least as a heavily manipulated image.
