Artificial Intelligence (AI) and stock photography

The internet is now full with videos and texts about the doom and gloom of photography in the face of Artificial Intelligence. Ethical and legal considerations about this new techmology are also being widely discussed. I have followed the development of AI for a while, particularly its impacts on the visual arts. This has caused me to add AI to my list of anxieties about the future of my modest earnings in photography, and that is what this post is about.

My photography income comes mostly from licensing images through stock photography agencies such as Shutterstock, Adobe Stock, Getty etc. I also sell prints, but that’s a subject for another post. Below I discuss the challenges AI might pose to stock photographers. But first, a short intro to AI and photography.

AI in photography

Whether we like it or not, AI is present in modern cameras and cell phones and allows us to track subjects, recognize the eyes of a person, calculate white balance, just to mention a few examples. In post-processing, it allows us to isolate subjects, perform selective changes in an image, replace the sky, get rid of unwanted elements like electricity poles, apply filters (presets) etc.

Text-to-Image generators use AI to produce images without a camera. In this case, all the user needs to do to obtain an image is to input a brief description of his vision into a computer program. If you are not familiar with the concept, I have found a link to a video explaining it (certainly better than I can): The text-to-image revolution, explained.

Both types of AI make it possible for the public to generate their own content to illustrate a product, an idea, or a concept without the help of a professional photographer. Think of a blog post, for example. Almost anyone these days can use a cell phone to take a picture illustrating a concept, then modify it with a few clicks and get an image that is good enough to post. And pretty soon it will be possible for almost anyone to obtain an image from a Text-to-Image generator, based on a few prompts.

AI and Stock Photography

I was delighted to read Megan Paetzhold’s article on the nymag “Will DALL-E the AI Artist Take My Job?“ because the answer for her is “not just yet”. She is a market consultant whose job is to “find or produce the visual elements that accompany New York Magazine articles”. After experimenting with DALL-E (the AI image generator from Elon Musk’s OpenAI consortium), Megan concluded that she outperforms AI in her work. I, on the other hand, provide photographs for that “Getty Images” subscription she uses to develop her concepts, and I can see how many of my photographs in the Getty and other databases will become obsolete very shortly. To get an idea of how good this technology already is, take a look at the DALL-E Instagram feed.

Stock Photo Secrets has just released an article “AI Generated Images: The Next Big Thing in Stock Media” about the moves stock agencies are making towards embracing AI to generate user content from a word prompt. These moves will most likely sharply decrease the number of photographs that are licensed, particularly those that represent broad concepts. For example, a relaxing place, a hot day, a healthy meal, a dark forest, a beautiful sunrise or sunset, a cup of coffee sitting beside a computer, these images can all be created in a few seconds using current AI technology.

I believe that many users will eventually bypass the stock agencies altogether, and create their own concept-driven images directly from open and free text-to-image generators. For instance, DALL-E has just brought on board 1k users from its waitlist. It gives them free credits to generate, edit or create variations of images, and an opportunity to buy more credits. All images created with DALL-E come with full usage rights, including the rights to reprint, sell, and merchandise the content.

A portion of my portfolio in stock agencies is dedicated to broad concepts. I also have three other types of images there: 1) lifestyle (real people doing stuff), 2) editorial photos of events, places and people, and 3) landscape photos of real places; 4) environmental conservation photos ( wildlife, rare flowers, fire damage, landslides etc).

The most lucrative stock photography genre is lifestyle. Lifestyle images are also more difficult to create because they require models that will sign a release. Paying models or convincing your friends and family to be photographed and sign a release is often a challenge. AI can easily solve this problem by creating realistic representations of people who do not exist. vAisual has pioneered the field by offering a plethora of AI generated faces and makes it possible for users to create their own faces using their dataset.

How can stock photographers compete with AI?

We probably can’t. However, in my view, there are still two broad categories of stock images that AI-generated images cannot replace just yet: editorial photographs, and photographs of real places. Also, photos of rare wildlife and botanicals will probably be difficult for AI to generate in the short term, and there are always nature-oriented publications that will need them.

In news magazines and newspapers, the photos portraying real people, events or property are supposed to be pretty much “out of the camera”. Cropping, contrast, and exposure adjustments are generally all that can be applied to editorial images, since they are supposed to reflect reality. Royalty-free landscapes, and wildlife/floral images, on the other hand, can, and often are, modified to some extent: for example, sky replacements, or getting rid of distracting elements, are common practice. The final product is still, for the most part, the real thing. I suppose that a travel magazine will still want to publish a photo of the Yosemite Valley or Bikini Beach, not an abstraction of those places, and a nature publication will want environmental conservation photos which, somewhat like editorial photos, aren’t supposed to be abstractions or to have been significantly modified.

Another way to protect your earnings from the propagation of images generated by AI, at least for a while, is to produce videos, because the AI technology for motion is not quite ready yet for the competition. 

Questions for my followers:

1) If you are a stock photographer, do you agree that editorial photos (containing property and real people and aimed for news outlets), landscapes portraying real places, and environmental conservation photos will be, at least in the short-term, shielded from the competition with AI?

2) More broadly, in your opinion, how can photographers and other artists protect their earnings in face of the new AI developments?

Conclusion

For the time being, and given the information I have so far, I will concentrate on uploading editorial images, and photos of real places and environmental conservation subjects because I believe that those will not be replaced by AI-generated images, at least not any time soon. I will also try to upload more videos.

The featured image

The FEATURED IMAGE was created with a pinhole camera mounted on a DSLR. I had the model move her head up and down slowly when the shutter was open. A fun, simple image to create.

I gave a random online image generator, which shall remain nameless, the prompt “a woman looking desperate, with her hands on her face, black and white, blurred image” and I got the image below. Not quite my vision but I must confess that I did not invest too much time on refining it.

If you are interested in trying, here is a link to an article with five Text-to-image generators that are free to try.

______________________________

Wall Art Botanical Images

Wall Art Photography projects

Wall Art landscapes and miscellaneous

________________________________

Published by Alessandra Chaves

Photographer with a preference for nature photography in black and white and other abstractions.

13 thoughts on “Artificial Intelligence (AI) and stock photography

  1. Your featured image at the top of this post really taps into my own feelings about this and so many other things lately. Nice. Of course, before last year we were already seeing marketing of the term “AI” in various software tools to help us develop and improve our images. It was only a matter of time before it progressed exponentially to full creations.

    And this isn’t only happening with visual arts, the Chatbots and music-generating tools are generating just as much concern. I believe we need technical solutions to address technical concerns and have been trying to formulate my own thoughts about AI. Maybe I’ll do a blog post about it soon.

    As I also do some stock, I do think this will have an immediate impact on certain segments of that market, particularly the more “generic” photographs and graphic art. Maybe I am wrong, but I haven’t seen many good examples of very specific locations or subjects that have been good replacements created by AI. Then again, this whole thing is just growing so fast, it is perhaps just a matter of time again.

    Liked by 2 people

    1. Mark, I found your comment in the spam folder, another blogger called my attention to look there. I would like to read your thoughts about the matter. There’s already a lot of debate about and law suits over copyright issues. My best guess is that we will have to live with it and adapt the same way film was replaced by digital and cameras by cell phones and photographers by amateurs, photographs will be replaced by holograms soon. I don’t know. The world has left me behind.

      Liked by 2 people

  2. I did microstock for a while and IMHO, that horse is dead. Even old photos bring less than it’s worth – in terms of your time – to upload and keyword them. Stock agencies have destroyed the market by competing on nothing but price.

    And now I think they’re going to get what they deserve – their own market will be destroyed in turn. People who currently buy stock for commercial purposes will just use these generating tools themselves – or hand the job to student interns. The tools will evolve to produce images that look, oh yeah, “authentic” and all that.

    The long term issue, though, will be copyright. These so-called “AI” tools that are “trained” by existing photos are actually just copying images in new and clever ways that the developers think will avoid existing copyright laws. I think what they’re really doing is extracting lines, colors, shapes and textures – in other words, they’re storing “vectors” made from our bitwise images. Does that amount to illegal copying? Eventually, a court will decide.

    Liked by 2 people

  3. I haven’t done any stock photography, myself. It is something I’ve thought about doing when – I retire less for income than as a way to stay active and maybe allow for keeping up my equipment – I just never really thought there was much money to be had there (I may be wrong there but I hadn’t yet done the research, so just an assumption).

    For now, it does seem the the fakeness in images are somewhat detectable. I’ve played around with Jasper and couldn’t come up with anything I didn’t look at for a while without sensing something off about it images. That could be my prompts, or there may be other, better tools out there.

    The whole kerfuffle over AI (chatGPT, Jasper, etc) has me somewhat upset, though, less for it’s income stealing aspect than for it’s ability to create convincing false narratives and pictures. Given the willingness we’ve seen lately we’ve seen exhibited by people to believe things so obviously untrue, in the face of so much evidence to the contrary, and equally their willingness to accept obvious fiction as fact – what will this sort of AI do for society when it is used (and it will be) to produce ‘credible proof’ of those falsehoods?

    Liked by 1 person

    1. Well, for those who started in micro stock more that’s 5-6 years ago, there’s still money to be made because you’re likely to have many assets that come first in the search engine searches. Those who start now May have more luck w video. I would not start now. I sent you a link through Instagram of a profile that’s based ok AI. Those images to me look very pretty and are probably the way forward fir a profitable print selling business…. People believe in the most outrageous conspiracy theories without the help of any image… just imagine if images were provided. The other side of it is, my grandfather never believed man stepped on the moon. He thought it was all fake photos and movies.

      Liked by 2 people

      1. Thanks for the link. I actually saw it before I saw this reply and didn’t know at the time the images were based on AI. Interesting. Coincidentally, I read an article in this morning’s Financial Times Weekend (about an hour after I commented on your post) about Getty Images suing OpenAI for improperly using its images to train DALL-E. Getty has allowed, with proper licensing, other AI developers to use their images for training, but it seems OpenAI and Stable Diffusion, another developer, used the images without proper licensing. These are going to be a very interesting next few years.

        Liked by 2 people

  4. You’ve given us a lot to think about. I watched the video and the bonus video that came up afterward. You raise the question of what kinds of photographs AI won’t be able to replace. One answer is actual occurrences at the time they happened. That includes things like a sporting event, a meeting of an organization, a politician’s visit with constituents, a car crash, a power plant under construction, children receiving awards at a school (except in northern Virginia). I assume those are the kinds of things you include in your category of editorial photographs. It also seems plausible to add, as you did, real places as they looked at specific times.

    I wonder whether legislatures will eventually pass laws dictating that photographs used for commercial purposes must be labeled as real, computer-generated, or a combination of the two.

    It’s interesting that that AI-generated portrait you closed with doesn’t have the woman’s hands on her face, even though that’s what you stipulated in your prompt.

    Liked by 2 people

    1. “One answer is actual occurrences at the time they happened.” I totally agree! Lol except in Northern VA. I think it will be the case that disclosure about the nature of the image will be mandatory and necessary.
      I have heard that hands are a particular challenge for AI at the moment. Not sure why.

      Liked by 1 person

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: