Stone pyramid and starry sky
Lens-Artists,  Photographic techniques,  Themed galleries

Gallery: AI versus human

Stephen Hawking

AI technologies can be of great service to humanity and all countries can benefit from them, but also raise fundamental ethical concerns, for instance regarding the biases they can embed and exacerbate, potentially resulting in discrimination, inequality, digital divides, exclusion and a threat to cultural, social and biological diversity and social or economic divides.

UNESCO

So what does all this mean for us as photographers? At its most basic level AI offers us new ways to improve our images. I’ve shown how previously I sometimes use Topaz AI to sharpen and/or resize images, and it does a great job of this. Photos I might otherwise have considered unusable become perfectly acceptable, even good!

But what about using AI to change an image so much that it no longer looks like the subject you photographed? Turning a blue or grey sky to a sunset? Removing an ugly building and replacing it with a tree? Adding objects, animals, people to a scene to create additional interest? While all of this may result it great images and be fun to do, personally I feel those of us who do so should always ‘fess up when sharing the results of our experimentations.

I do sometimes enjoy playing around with sky replacements in Luminar Neo and my feature photo from Teotihuacan in Mexico is clearly one such. I rather like the effect!

To date however I haven’t experimented at all with the creation of an image from scratch. But this week John asks us to consider the implications of AI for our photography for the Lens Artists challenge. Rather than revisit the topic of more subtle AI editing I decided to have fun with image creation, prompted by the last pair of shots in his post. Below I’m sharing a few of my own shots, paired with an AI generated image. For all of them I used CoPilot with a prompt based on a loose description of my own photo.

I’d love to hear what you think of the results! In my own view, sometimes it got it very wrong, other times created an interesting image but never realistic enough to appeal to me or for me to want to claim it as my own. My conclusion? I’ll stick to Topaz AI for sharpening etc., and may sometimes consider sky replacements or other more subtle AI edits, but I can’t see that I’ll find a use for this sort of image creation from scratch. I’ll leave that to those who have no interest in taking their own photos perhaps!

CoPilot produced all these images as squares. I have cropped some to a landscape format and added the white frame but made no other changes to them.

River with a bridge and tall skyscraper beyond

My image: the Thames with Shard, London

PROMPT: Bridge over the Thames with the Shard beyond

RESULT

Ornate bridge with red bus and two skyscrapers

Looking up at a white stone church with several domes

PROMPT: Trees partly obscuring the Sacré Coeur in Paris

RESULT

Domed building surrounded by bare trees

Landscape photo with a small house with red roof in front of snow covered mountains
Small house surrounded by snowy mountains

Two elephants in a river with sun setting behind
Two elephants in a river at sunset

Man playing the accordian

PROMPT: Old man playing an accordion on a quiet city street in black and white

RESULT

Man playing an accordion

73 Comments

  • Annie Berger

    Loved this post for its take on an altogether different subject but one we should all be aware of, Sarah. BTW – loved your feature photo with the starry sky. Also, appreciated the perfect quote by Hawking for this post. Hadn’t seen other AI-generated photos before now so these prompts were a tad alarming if that’s the future!

    • Sarah Wilkie

      Thank you Annie 🙂 I was pleased how that photo of Teotihuacan came out – it may not be ‘real’ but it is rather effective. The fully AI ones were much less successful but I suspect it will become cleverer and more sophisticated with time so maybe you’re right to be a little alarmed!

  • equinoxio21

    PS. I’ve probably mentioned using AI (Chatgpt) already, for translation from Spanish to English. Saves me half the time. I still edit every sentence.
    I just think it is a powerful tool to be used. The human using it makes the final decision.

      • equinoxio21

        Way better. Google translate is bad, works on a word by word basis. Chatgpt reconstructs sentences in a correct fashion. All I have to do is change a word for another, more like, hmmm, i would rather use another word, as more “my” word. Or modify a phrase more to my liking. With Google I had to practically rewrite everything. Chatgpt? Saves me at least 50% of the time I would spent retranslating or using Google. (Which is why you guys now have new(old) stories more frequently. Actually just posted one. (First half)
        Cheers

  • equinoxio21

    The street musician is very good.
    A friend of my youngest daughter has a company that’s hired by studios to “clean up” sequences in films, an ad, irrelevant people in a shot. They use specific programmes for that. He’s no exploring AI full time…

  • bluebrightly

    I agree, the last Ai image comes closest to acceptable but the others are pure fantasy, which wasn’t what you were looking for in this case.
    I don’t know if you know the blogger Dirk from the Netherlands. He experimented more extensively and got some surprising results. Not that I’m comfortable with the idea, because there’s no question that people will NOT ‘fess up. Here’s an example: https://fotografiemijnlangleven.wordpress.com/2023/11/16/adobe-firefly-ai-tekst-naar-afbeelding-part-3/
    Ai is here to stay – it’s going to be interesting!

    • Sarah Wilkie

      Thank you for that link. I had a look and I can see he achieves more realistic results than these, as have other people I’ve seen who use Adobe. I don’t have full Photoshop, only Elements, and I wouldn’t get it just for access to Firefly, as I don’t see me using AI to create images from scratch in the future (although I guess we should never say never!) But I also noted he used ‘photorealism’ in his prompts where I just used ‘photo’, so I’m wondering if that would make any difference…?

      • bluebrightly

        It might – his search terms seemed very specific. That being said, I don’t think either you or I am interested in going too much further down that road. 🙂

  • leightontravels

    The downright surrealism that often comes with AI is worrying / depressing / laughable depending on your perspective. Seeing the double Shard and that ludicrous bridge is so similar to the kind of images AI treated us to recently while we were creating a poster to advertise a cultural course for our online English school. All we wanted was a standard Thames scape image but what we received was some manner of interplanetary London with shimmering stars and comets, double decker buses driving through the water and not one but two Big Bens. I agree with you on the old man being the best of the bunch here.

    • Sarah Wilkie

      Wow, that image you generated sounds even more fantastical than any of these! I’ve seen examples from other people that are rather more realistic but very few that could have convinced the person in question took the photo (and of course, all of them were very clear about which they took and which were AI)

    • Sarah Wilkie

      Thanks so much Ritva 😊 Yes, I plan to stick to my own photos although I’m happy to use AI to improve sharpness and reduce noise 🙂

  • Sofia Alves

    The happiest elephants I’ve ever seen 🙂
    Great post, Sarah. I liked you started your prompts from original photos. It never comes close to what you thought. The last one is by far the less strange but even then there’s an illustration feel to it, not photo.

  • grandmisadventures

    The quote is very thought provoking as we are full on in the world of AI and now we’re trying to figure out what to do with it and where to go with it. As far as pictures go though- your AI pictures are really fantastic. I love how you put both a picture that you had taken and a picture that you had created and the different perspectives they give.

  • Philo

    As it stands, I feel AI doing reasonable job, not to my total satisfaction.
    Your original images are for superior ti ai generated ones.
    This is my personal opinion

  • Leya

    Efficient, Sarah! Very easy to see the differences with comparisons. And I agree – the last one is quite good, the best one. I too found that AI overdid my prompts, but it is not far away now…

    • Sarah Wilkie

      Thank you Ann-Christine 🙂 Yes, CoPilot definitely overdid my prompts but I’ve seen more successful attempts in other people’s posts using different programs – I guess some have evolved more quickly.

  • Tanja

    The last one is the only that looks real. I am afraid that soon it will be very difficult to discern human photography from AI photography. And whether the photo captures a real moment or is it altered…..

    • Sarah Wilkie

      Yes – normally I get a variety of opinions on my posts (different favourites, different views on editing etc.) but not this one! I agree about the Thames bridge, it’s very weird 😀

  • carabeinsplash

    Great idea Sarah! Clearly the AI examples couldn’t be passed off as photos because they look like illustrations. The old man looks more realistic but just too uniform and perfect. Realism includes flaws. I liked your actual photos so much better!

  • wetanddustyroads

    That’s a very good comparison Sarah! However, I can say with certainty that I prefer your photos to those of AI instead. The old man with the accordion is also the closest that looks like a “real” photo … the rest look like … well, like AI.

    • Sarah Wilkie

      Thank you – I think the general consensus is that only the old man image would convince anyone that it’s a real photo and even then there are giveaways such as his too perfect skin!

  • bushboy

    Originals vs AI was a good presentation. Good on you for having a go to get results that are less than satisfactory unless you are designing covers for fantasy novels.
    I am on team accordion player which is close to what you ordered. If AI was a restaurant I would have sent every dish back

    • Sarah Wilkie

      Haha, I love that comparison with a restaurant 🤣 To be honest, some of its earlier attempts at my prompt were even worse than these and I did indeed ‘send them back’!

  • thehungrytravellers.blog

    I am such a technology dinosaur that I only half understand this post 😂. On another subject Sarah, do you want to email me (at thehungrytravellers.blog@hotmail.com….) with some dates when we might be able to have our long awaited meet up. Our diaries are a little bit complicated by the hospital issue but we could at least perhaps pencil something in. Let’s say available dates between now and 10th July…?

  • Tina Schell

    Well done Sarah. I liked the way you used your own image for comparison and included the prompt used for AI. I agree the AI creations are quite fantasy-like except for the last which could definitely pass as a photo. It’s a strange new world for sure!!

  • Rose

    What a delightful comparison. I kind of like the AI versions of the Thames bridge, and the red roof home at the bottom of the mountains. It is disconcerting to think that someone may try to pass off AI images as reality.

    • Sarah Wilkie

      Thank you Rose 🙂 The Thames one is too fantastical for me but the red roof house has something of the charm of an old-fashioned children’s story book I think.

  • kzmcb

    Very interesting post, Sarah. I’m not convinced that AI is for me and I’ve felt the same when using it for stories and essays.

    • Sarah Wilkie

      Thanks Jo 🙂 The accordionist is definitely the best although as others have pointed out his skin isn’t convincing – wrinkled in parts but too smooth in others.

  • Wind Kisses

    I love the way you organized your thoughts for this one Sarah. Using your photos first and then giving an AI prompt brought strength to your own photos. Some of the AI interpretations were science fiction looking. And I will agree, the man with the accordion is the best but he still looks a little cartoonish. I knew I would like your post. I did.

    • Sarah Wilkie

      Thanks so much Donna 😊 I was stuck for ideas for a while, then this came to me and was quite quick to do – necessary as I’m just back from a trip!

  • Anne Sandler

    I preferred all your originals. The AI images looked like typical AI–all except the last. This image of the old man looked like something you would take. I agree that you need to let people know if you’re posting an AI image. I’m not talking about the usual edits or sky replacements. I’m having trouble figuring out how to do AI. I’m not technical at all and having a BAD cold is not helping me concentrate. I tried doing sky replacements but the lighting has to be followed through. Again, I don’t know how to do that. So I guess whatever I post will be mine!

    • Sarah Wilkie

      Thanks so much Anne 🙂 I feel even sky replacements should be ‘declared’ most of the time, although if you’re doing a subtle change that is still close to what you saw (e.g. if a bright sky has turned out too washed out) then it’s OK. As to lighting, some programs do that for you, to some extent at least. But try Googling ‘CoPilot’ as it’s free and easy to use, although as you can see the results are bizarre!

  • Sue

    Well! A most interesting experiment, Sarah….that AI programme that you were using was truly rooted in fantasy land! However, the last image wasn’t bad But your photos are what I would go for

  • Anonymous

    Your own images are far superior to the AI-generated examples. Truly, that black-and-white could have easily persuaded me that it started with a real image and was edited for a mystical effect. Only the face appears to be “illustrated” rather than photography.

    Now, I think you ought to submit that AI bridge to the powers that be. I’d like to see that “for real.” >grin<

    All in all, it's been interesting to see people's attempts and read their thoughts on the new world of photographic AI.

    • Sarah Wilkie

      Thank you ‘anonymous’ (I wish I knew who you were!) 😀 I agree that B&W shot could be an edited ‘real’ photo. I’m not too sure about that bridge however!

  • Egídio

    Sarah, your experiments with AI are fascinating. Somehow, AI overdid your prompts, e.g., doubling the buildings, etc. I agree that the final image produced the best results — even if the man’s face skin gives it away. Great post!

    • Sarah Wilkie

      Haha yes, I get the impression AI can’t count too well – several of my ‘two elephant’ prompts resulted in three or even more! And I agree about the man’s skin – wrinkled in places, totally smooth in others.

  • Monkey's Tale

    As I’ve said previously I don’t like AI generated pictures especially those that are vastly different from reality. Your’s are a great example of why. The only one that is near normal is, as you’ve said, the old man. Maggie

  • Terri Webster Schrandt

    Playing around with AI generated images is fun if not a bit frustrating. But the more we do this the more AI will learn. Of course your images are always amazing!

    I played around with AI gen with WordPress to help define an image for my blog site. I found one, but i had to delete a LOT from my media files.

    As photographers, we have our own eye, not an A eye. 🤣

  • Ingrid

    Yes, only the last image could pass as a real photo. The rest are clearly in a fantasy realm. I much prefer your real photographs over anything AI created. This was definitely an entertaining experiment. It’ll be interesting to see how AI improves over time and its overall impact. I have mixed emotions on the topic.

    • Sarah Wilkie

      Thanks Ingrid – I agree, an entertaining exercise but at the moment at least not one I’m keen to repeat. However, let’s see how AI develops …

  • margaret21

    That’s an interesting experiment. And for sure, only the last would pass muster as an actual photo. I can’t think that you, or any of the other photographers who regularly post on Lens-Artists could be guilty of such schmaltzy scenes. I get the feeling that AI must be powered by Clinton Cards or similar!

    • Sarah Wilkie

      I think it depends which AI program you use as I’ve seen more convincing results in some other posts. But yes, schmaltzy is definitely the word for most of these!

  • JohnRH

    GREAT examples and comparisons. For most of yours, AI seems to be in ‘fantasy’ mode. Is your header photo of the stars and ruins real, or Memorex?

    • Sarah Wilkie

      Yes, the CoPilot default seems to be fantasy, I couldn’t persuade it to do anything else in most cases! The header photo is as I explained in the text – one of my images with the sky replaced in Luminar Neo 🙂

  • Teresa

    Yes, for sure the last one is the best one of all. The others are also amazing if you don’t take into consideration the reality and placement of buildings and trees. You did well, Sarah. But of course your photos are still the one i’d choose for sure.

Do share your thoughts, I'd love to hear from you! And please include your name in case WP marks you 'anonymous' - thank you