I have been working with artificial intelligence (AI) as a visual art form for a couple of months as of this date. By that I specifically mean text-to-image generative AI. Now AI can be used in so many different fields – architecture, graphic design, fiction and non-fiction writing, music, and so on. I am not concerned with those, but only with the perception of text-to-image generative AI as it relates to the visual arts world.
It all started with two recent AI software advances for photographers. The first was Topaz AI and the second was when Adobe started advertising the new generative AI feature for Photoshop subscribers. As an enthusiastic photographer for most of my life, I decided to try these new tools and quickly realized what they could do to improve my photos, mainly for me by eliminating unwanted "things" in existing photos – people in the wrong places, poor quality and undramatic skies, poor lens focus, excessive noise, and so on. Plus I could also extend the canvas of my images with using generative AI input.
So here is the first question. Does my use of AI for these purposes take my photos out of their place as pieces of original art by me as a photographer? My answer is a firm NO, because I could also have done the same improvements via Photoshop but the process would have been much slower and the results no better. The only caveat I would put on this conclusion would be that if I intend to sell the finished image then ethically I must declare it to be partially the result of AI enhancement. To me there should be no argument that the image is still my own art.
Next came true text-to-image AI. Once into improving my photos by AI, I decided to try full-on AI. After a few days of seeing amazing results with my prompts, I was hooked. I started creating a lot of generative images, usually on similar topics to those that I enjoy doing in my photography. Some of these included realism, classic pinups, historical situations, and famous femmes fatales, for starters. I have since been offering some of these images on my Fine Art America website for sale (https://doug-matthews.pixels.com). The reaction by viewers has been very positive.
Now here is the second critical question. Can the images I create be considered true, copyrightable art? Again, my answer is a resounding YES. Here is why.
First, any given image is generated by a written prompt, the creation of which is my own. The probability of achieving a successful image will be entirely dependent on my own creative writing skill. Because of this and the way that AI works, no image will ever be duplicated in its entirety, so each one is indeed unique and a result of my singular input. Essentially, the written prompt is to AI what a paintbrush is to a painter or point-of-view and shutter click are to a photographer.
Second, any given image is an amalgam of the knowledge available to the generating software, further enhanced by certain proprietary inputs (e.g. upscaling, style like fantasy or realism, etc.). To suggest that this is theft of intellectual property is ridiculous. It is no more intellectual theft than, for example creating a pinup in a classic style using my own model in a photo session and my own advanced knowledge of Photoshop in post production.
There is no question that text-to-image generative AI moves the dial with respect to creative visual art. This seems to be what artists and others are so afraid of, if the truth be told. They are worried that AI will usurp their positions in the arts world. Yes, there will be a few who will suffer. Models and graphic artists are two fields that probably will. But AI is an amazing new genre of art. We should accept it and adapt to it. Three stipulations that I believe are absolutely necessary, though, and those are truthfulness, honesty, and integrity. By this I mean that all AI artists must be brutally honest and label their art as clearly AI – no attempting to pass it off as photography. Also, AI must be its own genre or medium, and not be force-fit into others. It can certainly have an inherent style (e.g., impressionism, realism, cubism, etc.) but those should be sub-classifications within the genre not without.
To finish this post, I'll illustrate the above discussion with recent examples of my own AI work.
For this first image, I wanted to portray a realistic circus theme. The final text prompt that got me to the first part of this image was,
"A circus ringmaster seen in profile, watches a big top tent go up. A female trapeze artist and a clown can be seen in the background."
Well, that didn't work out so well. It only produced the ringmaster. After this I tried the following,
"A sad clown and a happy female trapeze artist stand together."
Results were not very good so I isolated the trapeze artist in Photoshop. The image was still a little boring so I added an elephant with the prompt "circus elephant on hind legs." This time I just could not get the elephant on his hind legs and instead got a mish-mash of body parts.
I further isolated the elephant, resized him, and adjusted lighting within Photoshop as well as combining all three images in Photoshop.
My point in relating all this is that final images rarely come out of an AI prompt on the first try and most require further manipulation. Could the realistic images of people be taken from elsewhere? Possibly. The thing is, I have no way of knowing and the program I use, called Leonardo AI, tells me I can use any of the images I create with no compunction, and that they are copyrightable to me. The jury is really still out on this one.
This second image was intended to portray an aviator at the very first big aviation show in Reims, France in 1909, my personal visualization of the event. It is much more photorealistic than the first image. Again, I used different generating software within Leonardo AI. It was one of the results of the following text prompt that resulted in 47 generations, from which I chose this one:
"An early intrepid aviator attired in leather flying cap, flying goggles, a black leather jacket, and smoking a cigarette. He has a large moustache. He looks very confident. He is in three quarters view and looks at an airfield in the distant background. An old biplane is flying over the airfield out of focus."
As you can see, it missed the cigarette part, although the pilot looks realistic. Also the initial image was smaller and had ridiculous depictions of biplanes in the background. I reverted to layers in Photoshop to put in the biplanes that were taken from old postcards I own, along with the tower in the background. I spent a fair bit of time on blending the layers properly to achieve the final result. I have no qualms with saying this is an original piece of art created by me and copyrightable to me. From what I can tell, nothing was "stolen" or "copied" directly from anyone else's art.
What do you think? Does AI threaten you as an artist?
Commentaires