I’ve noticed something lately, and maybe you’ve noticed it, too. There are some Fujifilm recipe creators using AI to create the sample “photos” that illustrate what their recipes will look like. Instead of using their cameras to create actual pictures with the recipes that they supposedly made, they have AI do it for them. Whether you love or hate AI, I think we can all agree that not using real photographs to illustrate what a recipe will produce is just wrong. It’s dishonest and deceptive at the very least.
To illustrate the problem, I had AI create the “picture” at the top-left using nothing more than the Fujifilm Recipe parameters—in this case, Retro Slide—plus a brief explanation of what the subject should be. The photo at the top-right is an actual image created using the Recipe on my Fujifilm X-E5 camera. You can clearly see that the AI image is not a good representation of what the Retro Slide Recipe will produce. I did this a bunch of times with various Recipes, and not once did the AI image look like a good example of the Recipe aesthetics. Some were better than others, but none were particularly close matches.
Why would someone do this? Well, maybe they’re not an actual photographer. Maybe they don’t even own a Fujifim camera. Maybe they are well connected to the camera industry and even work for Fujifilm—a true story, by the way (yikes!!). I don’t know why someone would do this, other than AI is a cheap and easy shortcut that removes the prerequisite work that used to be essential. It seems like the minimum effort should be to actually use the recipe in an actual camera to create real pictures, but apparently not anymore.
Above left: AI-generated sample image for the Fujifilm Negative Recipe; Above right: an actual image made using the Fujifilm Negative Recipe on a Fujifilm X-T5.
Now if the example “photo” is AI, the recipe itself was likely created by AI, too. While AI has gotten better at this task (I test it every now and then to see the progress), it is still a little ways off from being a good tool for making recipes. These programs seem to know a lot more about Fujifilm camera settings than just two years ago, but they still don’t have a great understanding of precisely the real-world implications of each parameter adjustment. It’s also very inconsistent: ask AI the same prompt using 10 different accounts, and you will get 10 different answers. How can you know which is the most correct and which one is the least? It doesn’t mean that they’re all bad, but it does mean that you need to shoot with it and test it and see if it is good, because it very well might not be. AI is a seemingly magical tool; however, it should be viewed as low-trust, since it is so inconsistent and gets things wrong often enough. As someone once told me: AI is much better than a complete novice, but nowhere near as good as an expert.
When I create Recipes, I test them, refine them, test them again, refine them again, and test some more. This can go on and on, sometimes for weeks or occasionally even months. With a few exceptions, I would estimate that each Recipe I make requires a minimum of 100 exposures before it’s published (sometimes a lot more than that). While it does take some effort, I feel like it’s completely worthwhile to ensure that it meets my standards. I couldn’t imagine publishing a Recipe without at least testing it once, yet that seems to be happening.
I make Fujifilm Recipes so that I and those who use them get out with our cameras to create photographs, real photographs. Obviously we could just sit at our computers all day and prompt AI to do it for us, but that would be a massive loss. I’ll be at Fujikina NYC this coming weekend (I hope to see you there!), and one theme of the event is creating real photos in the real world along with real people. Real cameras. Real places. Real pictures. Real community. This is what photography should be about. This is what Film Simulation Recipes should be leading us towards. AI wants to pull us in the other direction, and that’s not positive. So let’s grab our cameras and create actual photographs that have real meaning for us.
That’s disheartening to hear. I avoid the use of AI in all forms; I absolutely don’t want it anywhere near my photography workflow.
I agree! I try to avoid it whenever practical, and I definitely don’t want it anywhere near my photography if it can be helped.
Sometime I use AI to evaluate the “mood” of my photo, and quite often noticed me some points that I did not considered. Anyway, I don’t want anything to do with creating or editing my photos. Creating random recipes with AI is also an interesting game
AI can be useful sometimes. I would be curious if you asked it to evaluate the same photo using several different accounts, would it give you the same answer?
Yes, I use different AI model and I get different detailed answer while the “general” impression is about the same
I think it’s ok to be open about the fact that AI is horrible, nobody asked for it and it makes everything worse
We probably need to be more open about how it’s likely going to turn out very, very bad for humans.
Even if AI could do this stuff perfectly I wouldn’t use it or like other people using it. The point of anything creative isn’t just the end result, it’s the fact that another human has created it and chosen to share it with other people.
100% agree.
Artificial ? Yes!
Intelligent? I have my doubts….
It’s a lot more artificial than intelligent…
I learned only today that there is such a thing as the IPTC Data mining permissions field: a meta data field in which you can make clear that your image may not be used for (Generative) AI/ML training, when you post it online or to social media. This does of course depend on the chosen medium not stripping all metadata from your images, like Instagram does…
I think the data has already been mined. ChatGPT admitted to me that it trained on my website. I didn’t agree to that, but the argument is that it is “publicly available”. I think the AI companies took the philosophy of it’s easier to ask for forgiveness than permission (which is unethical in my book). The damage has already been done.
I would ask, if it is even permissible by law, that the AI models crawl your Website for content and knowledge, if in your terms of use of your website you do not allow the data be used by AI. That, however, I guess, is a tricky thing, as you permit human users to use the recipes, you created.
What I think though, is, that this is a very important question to think about before more damage can be done.
When I think back, I used to see the Chinese go around trade fairs, and buying products, so they could copy them. That is the past, they have learned a lot and got good and innovative.
Now, it seems to me, that role of wandering around, and copying things has been taken over by all the AI models.
It’s going to take many years for the lawmakers and courts to hash everything out. Maybe by 2030 much of the legal stuff will be settled, and by 2035 nearly all of it will be settled. My guess (and this is just a guess) is that when all that happens, it will be determined that most or all of the AI companies illegally trained their programs. We’re already seeing that a little, but the depth of it will be much more understood. Of course, by that point the damage will have already been done, and there won’t be anything anyone can do about it. It is what it is, so to speak. In a sense, anyone who has ever published some creative original work (of any sort) will have had an aspect about it “stolen” from them in an effort to use it against them so as to put them out of business by replacing them. That’s just our new reality, our new paradigm. I think the incentive to create and share original art is being stripped away; however, what I think will happen is that art will move off of the internet as much as possible. Real, tangible pieces in real galleries/exhibits/stores that have never touched the internet is where it will move to. Artists will find ways to prevent AI from copying them, and people will yearn for things that are real and true and not artificial. That’s my opinion, anyway.
You are approaching it completely wrong. The ai did exactly what you asked it to do. IT CREATED AN IMAGE from parameters you gave it. Give it a real IMAGE you created and the ACTUAL FP1 file for the recipe and ask it to do the changes on the file. You will be surprised. ChatGPT has native abilities from Adobe.
But if you have the real photo, why would someone need to make a fake version of it? It would be like having Alaskan King Crab and some alternative crab meat in front of you—the latter might look and taste pretty similar, but you’ll always go for the former as it’s obviously superior. If someone has a real photo they don’t need to ask AI to make a fake one. The obvious conclusion is that the person doesn’t have a real photo in the first place. Many of the recipes with AI example “photos” found across the internet/app-store won’t produce results similar to the AI image, it’s so apparent. Whether ChatGPT has the ability to be more accurate to a picture is irrelevant, because 1) that’s not what is happening and 2) even if it is, ChatGPT images are still fake, they’re cartoons. As photographers, we should want nothing to do with that.
More slop of another kind. Some people prefer cheating, some people prefer stealing, some do honest work. It funny because people who steal or say seek drugs make less than if they did real work, and then they go to jail when they get caught. I guess it’s what you know. I’ve been getting recent version of chatGPT to critique my photos and suggest edits and crop and it’s quite good. I’ve only done 3 or 4 photos and it will take 50 to 100 to see how it really does. I have a paid subscription if that maters. With respect to critiques, it looks like chatGPT is much better than our photography club members who critique photos for competitions. It is at the very least much more skilled at telling us what is good or bad about a photo than we are. If it’s really competent as opposed to bulshiting, then we should start having members read chatGPT critiques as we would learn more. Really we are not very good at critiquing photos. Jury is still out.
I would be hesitant to use AI to critique for several reasons. First, it will likely give you different answers if you asked it from different accounts. Which critique is “right” and which one isn’t? Second, it likely relies on “photography rules” that are actually nonsense. For example, someone once told me that photography rules are meant to ensure consistently good photograph but often stand in the way of great photographs. AI can’t identify a great photograph unless people post to the internet that a photo was great; instead, it might identify it as problematic because it didn’t follow the “rules” that people perpetuate across the web.
Personally I don’t use AI to get a verdict, but to gain a different perspective that makes me think. The more varied the perspectives, the more I have to reflect on; not truths, but pointers. AI can’t tell beauty from ugliness or right from wrong because it simply has no form of awareness. We’re the ones who “lend” it our awareness by interpreting its responses, which, I repeat, make sense to us, not to AI. I make the final judgment by myself; at the end, even with AI’s input, I have to like my photos. But I say that as an amateur photographer who earns a living in other ways (ironically, including by implementing AI management systems).