AI doesn't have to be evil

We frequently portray AI as being evil and the beginning of the end for "Photography as We Know It."

Here's a shot that'll be in a brochure for a local Mexican restaurant. The place is decorated in an explosion of bright, saturated primary and secondary colors.

An image of a bucket of colorful lollipops for niños (and young-at-heart adults!).

It's under a table. I didn't appreciate how dark it was under there and had left the ISO at 100, resulting in a 1/17 sec shutter. Way too slow for the IBIS as I was leaning over and unsteady, with shaky hands.

As is, it doesn't carry off the look I wanted.

Topaz Sharpen AI to the rescue. A few seconds. No fuss, no muss. No drama. No artifacts. Auto settings simply made the shot look rock solid and accurately-focused. Extremely useable.

View at Original Size.





GFX 100S, GF 45/2.8, f/4.5

--
Rich
"That's like, just your opinion, man." ;-)

If we dig through the archives I bet we can find a thread from a few decades ago titled "PhotoShop doesn't have to be evil".

In all seriousness, I think it's important to distinguish the many different kinds of "AI".

Tools like denoise and super resolution are not in the same category as generative AI that can create entire images (or sections of images).

We're all going to have to put up our own guardrails because the industry sure isn't going to do it for us. I will use AI-based denoise tools. It seems to have a very light touch.

I don't like super resolution in Lightroom because there's more invention than makes me comfortable, and the results can be ugly and weird.

I avoid generative AI tools completely in post-processing. This isn't a moral judgement. Rather, images made using generative AI are not photographs in the way I think of the concept, so they don't interest me.

P.S. I think your client is going to love the image you made, and won't give a rat's behind that you used AI tools!

I don't believe AI is even necessary for this image. Deconvolution sharpening would be all that is needed. Try Photoshop's simple "smart sharpening" tool on images with soft focus or motion blur. It works wonders.

Wouldn't it be easier to just ask AI to generate a bucket of lollipops?

GonnaSnap said:

I don't believe AI is even necessary for this image. Deconvolution sharpening would be all that is needed. Try Photoshop's simple "smart sharpening" tool on images with soft focus or motion blur. It works wonders.

Click to expand...

The problem is motion blur. Deconvolution sharpening would not be the answer.

As I said, the program automatically detected the need and the result is everything I could want.

xtm said:

Wouldn't it be easier to just ask AI to generate a bucket of lollipops?

Click to expand...

🫩

Rich42 said:

We frequently portray AI as being evil and the beginning of the end for "Photography as We Know It."

Here's a shot that'll be in a brochure for a local Mexican restaurant. The place is decorated in an explosion of bright, saturated primary and secondary colors.

An image of a bucket of colorful lollipops for niños (and young-at-heart adults!).

It's under a table. I didn't appreciate how dark it was under there and had left the ISO at 100, resulting in a 1/17 sec shutter. Way too slow for the IBIS as I was leaning over and unsteady, with shaky hands.

As is, it doesn't carry off the look I wanted.

Topaz Sharpen AI to the rescue. A few seconds. No fuss, no muss. No drama. No artifacts. Auto settings simply made the shot look rock solid and accurately-focused. Extremely useable.

View at Original Size.





GFX 100S, GF 45/2.8, f/4.5

Click to expand...

I do agree and what people call AI is really ML and requires people to train the neural networks and get weight matrices for their intended function. So they can be only as good or evil as people can be. Which means we have to pay attention to how and what they are used for as I think we all must recognize as any human made tool can be used for good or evil. For example, ML will be used for military purposes and I can understand that. What I don't understand is the maniacal desire the have AI replace artists wholesale or these super gamer AI's. Do they think this will replace humans desire to make art or play games. I have questions given the areas it could be really helpful in.

xtm said:

Wouldn't it be easier to just ask AI to generate a bucket of lollipops?

Click to expand...

I asked ChatGPT to

"Create an image of a shiny metal pail of multi-colored, cellophane-wrapped lollipops with white stems."



It misunderstood the context of "multicolored." or I miscommunicated the idea. It's a great image!

But it's not the bucket of pops in the restaurant for which the brochure is being made.

Now if I could only actually find such lollipops as these and have the restaurant use them instead . . . !

--
Rich
"That's like, just your opinion, man." ;-)

Rob de Loe said:

I avoid generative AI tools completely in post-processing. This isn't a moral judgement. Rather, images made using generative AI are not photographs in the way I think of the concept, so they don't interest me.

Click to expand...

I recently came to the same conclusion after some internal debate. To me, a photograph is a photograph because it is a projection of a state of the real world at the time of capture, and generative AI by nature can never meet that criterion. We could call it an image perhaps, but to me it would not be a photograph.

That's not to say that an image with varying proportions of photograph and generative AI cannot be art, but consuming and producing such art does not interest me in the same way photography does (not to mention the mass theft and ethical problems surrounding most modern generative AI models).

I did have one dilemma, which was whether it would be unfair to treat generative AI in this way, but not traditional composites, content-aware fill, retouching and so on. My current thought is that in those cases, the manipulated content almost always comes from other photographs as well - but I can't help but wonder if the final modified photograph as a whole is less of a photograph than before? I'm sure some might disagree on this, especially seeing as such techniques are quite widely accepted and used.

--
https://warmskies-photography.carrd.co/

Rich42 said:

xtm said:

Wouldn't it be easier to just ask AI to generate a bucket of lollipops?

Click to expand...

I asked ChatGPT to

"Create an image of a shiny metal pail of multi-colored, cellophane-wrapped lollipops with white stems."



It misunderstood the context of "multicolored." or I miscommunicated the idea. It's a great image!

But it's not the bucket of pops in the restaurant for which the brochure is being made.

Now if I could only actually find such lollipops as these and have the restaurant use them instead . . . !

Click to expand...

Create an image of a shiny metal pail full of cellophane-wrapped lollipops with white stems, each lollipop a different, bright, saturated color . . .



It's good!

But . . . Nah! Looks fake.

--
Rich
"That's like, just your opinion, man." ;-)

Rob de Loe said:

If we dig through the archives I bet we can find a thread from a few decades ago titled "PhotoShop doesn't have to be evil".

In all seriousness, I think it's important to distinguish the many different kinds of "AI".

Tools like denoise and super resolution are not in the same category as generative AI that can create entire images (or sections of images).

We're all going to have to put up our own guardrails because the industry sure isn't going to do it for us. I will use AI-based denoise tools. It seems to have a very light touch.

I don't like super resolution in Lightroom because there's more invention than makes me comfortable, and the results can be ugly and weird.

I avoid generative AI tools completely in post-processing. This isn't a moral judgement. Rather, images made using generative AI are not photographs in the way I think of the concept, so they don't interest me.

P.S. I think your client is going to love the image you made, and won't give a rat's behind that you used AI tools!

Click to expand...

I've seen examples of an AI editing tool I don't have, but would like to have: the fancy tool that does cloning and then patches the background so that you would never know the object that had been removed had ever been there. I'm forever trying to remove small unwanted objects from a frame to tidy it up using manual cloning methods (with little skill).

An automated repair tool would be very useful if they ever come to open source editors.

aloli said:

Rob de Loe said:

I avoid generative AI tools completely in post-processing. This isn't a moral judgement. Rather, images made using generative AI are not photographs in the way I think of the concept, so they don't interest me.

Click to expand...

I recently came to the same conclusion after some internal debate. To me, a photograph is a photograph because it is a projection of a state of the real world at the time of capture, and generative AI by nature can never meet that criterion. We could call it an image perhaps, but to me it would not be a photograph.

Click to expand...

I'm with you. In fact, I would say this distinction is important because it gets at photography's "super power".

aloli said:

That's not to say that an image with varying proportions of photograph and generative AI cannot be art, but consuming and producing such art does not interest me in the same way photography does (not to mention the mass theft and ethical problems surrounding most modern generative AI models).

Click to expand...

I suspect the overwhelming majority of people won't care how the art was produced. I think that's unfortunate, but if I'm not in the minority yet, I will be soon. Most people just want to be entertained. AI slop that entertains will be fine for those folks.

aloli said:

I did have one dilemma, which was whether it would be unfair to treat generative AI in this way, but not traditional composites, content-aware fill, retouching and so on. My current thought is that in those cases, the manipulated content almost always comes from other photographs as well - but I can't help but wonder if the final modified photograph as a whole is less of a photograph than before? I'm sure some might disagree on this, especially seeing as such techniques are quite widely accepted and used.

Click to expand...

Photography has included composites since the very beginning. I've always thought of these as art pieces made with photographs.

Kelli Connell's Double Life series continues to amaze me. https://www.kelliconnell.com/new-page-5 The end product is a composite made with photographs, rather than a photograph, but that sits well with me because it's a human making rather than output from a generative AI tool. Sadly, her body of work in this project is large enough now that I suspect it would not be hard to train a generative AI tool to make new images that look like they belong in the series.

I wrote a bit about the implications of generative AI for photography a couple years ago. https://www.robdeloephotography.com/Pages/Generative-AI

I know that some people see a brave new world of image making using generative AI. I see the end of most kinds of photography that I value.

Rich42 said:

xtm said:

Wouldn't it be easier to just ask AI to generate a bucket of lollipops?

Click to expand...

I asked ChatGPT to

"Create an image of a shiny metal pail of multi-colored, cellophane-wrapped lollipops with white stems."



It misunderstood the context of "multicolored." or I miscommunicated the idea. It's a great image!

But it's not the bucket of pops in the restaurant for which the brochure is being made.

Now if I could only actually find such lollipops as these and have the restaurant use them instead . . . !

Click to expand...

The right hand stick is missing its lollipop.

--

aloli said:

Rob de Loe said:

I avoid generative AI tools completely in post-processing. This isn't a moral judgement. Rather, images made using generative AI are not photographs in the way I think of the concept, so they don't interest me.

Click to expand...

I recently came to the same conclusion after some internal debate. To me, a photograph is a photograph because it is a projection of a state of the real world at the time of capture,

Click to expand...

So Jerry Uelsmann and Mortensen didn't make photographs? How about Man Ray?

aloli said:

and generative AI by nature can never meet that criterion. We could call it an image perhaps, but to me it would not be a photograph.

That's not to say that an image with varying proportions of photograph and generative AI cannot be art, but consuming and producing such art does not interest me in the same way photography does (not to mention the mass theft and ethical problems surrounding most modern generative AI models).

I did have one dilemma, which was whether it would be unfair to treat generative AI in this way, but not traditional composites, content-aware fill, retouching and so on. My current thought is that in those cases, the manipulated content almost always comes from other photographs as well - but I can't help but wonder if the final modified photograph as a whole is less of a photograph than before? I'm sure some might disagree on this, especially seeing as such techniques are quite widely accepted and used.

Click to expand...

DavidMillier said:

Rob de Loe said:

If we dig through the archives I bet we can find a thread from a few decades ago titled "PhotoShop doesn't have to be evil".

In all seriousness, I think it's important to distinguish the many different kinds of "AI".

Tools like denoise and super resolution are not in the same category as generative AI that can create entire images (or sections of images).

We're all going to have to put up our own guardrails because the industry sure isn't going to do it for us. I will use AI-based denoise tools. It seems to have a very light touch.

I don't like super resolution in Lightroom because there's more invention than makes me comfortable, and the results can be ugly and weird.

I avoid generative AI tools completely in post-processing. This isn't a moral judgement. Rather, images made using generative AI are not photographs in the way I think of the concept, so they don't interest me.

P.S. I think your client is going to love the image you made, and won't give a rat's behind that you used AI tools!

Click to expand...

I've seen examples of an AI editing tool I don't have, but would like to have: the fancy tool that does cloning and then patches the background so that you would never know the object that had been removed had ever been there. I'm forever trying to remove small unwanted objects from a frame to tidy it up using manual cloning methods (with little skill).

An automated repair tool would be very useful if they ever come to open source editors.

Click to expand...

The automated repair tool in Lr is good and getting better all the time, but the copy/paste development settings doesn't work right with that tool, since it doesn't recompute the patches.

JimKasson said:

aloli said:

Rob de Loe said:

I avoid generative AI tools completely in post-processing. This isn't a moral judgement. Rather, images made using generative AI are not photographs in the way I think of the concept, so they don't interest me.

Click to expand...

I recently came to the same conclusion after some internal debate. To me, a photograph is a photograph because it is a projection of a state of the real world at the time of capture,

Click to expand...

So Jerry Uelsmann and Mortensen didn't make photographs? How about Man Ray?

Click to expand...

I do address this at the end, and my current stance is that I'm not entirely sure. Images where photography is the primary medium? Definitely. And they definitely made photographs as part of their creative process.

But I hesitate to accept that they are photographs (by my original definition) solely because these are well-known innovators in the world of photography, and because most people would agree that they are. Maybe I'll arrive at a clearer conclusion one day, but that conclusion would mostly serve to shape my own understanding of photographs and photography.

JimKasson said:

JimKasson said:

and generative AI by nature can never meet that criterion. We could call it an image perhaps, but to me it would not be a photograph.

That's not to say that an image with varying proportions of photograph and generative AI cannot be art, but consuming and producing such art does not interest me in the same way photography does (not to mention the mass theft and ethical problems surrounding most modern generative AI models).

I did have one dilemma, which was whether it would be unfair to treat generative AI in this way, but not traditional composites, content-aware fill, retouching and so on. My current thought is that in those cases, the manipulated content almost always comes from other photographs as well - but I can't help but wonder if the final modified photograph as a whole is less of a photograph than before? I'm sure some might disagree on this, especially seeing as such techniques are quite widely accepted and used.

Click to expand...

--
https://blog.kasson.com

Click to expand...

--
https://warmskies-photography.carrd.co/

Rich42 said:

Rich42 said:

xtm said:

Wouldn't it be easier to just ask AI to generate a bucket of lollipops?

Click to expand...

I asked ChatGPT to

"Create an image of a shiny metal pail of multi-colored, cellophane-wrapped lollipops with white stems."



It misunderstood the context of "multicolored." or I miscommunicated the idea. It's a great image!

But it's not the bucket of pops in the restaurant for which the brochure is being made.

Now if I could only actually find such lollipops as these and have the restaurant use them instead . . . !

Click to expand...

Create an image of a shiny metal pail full of cellophane-wrapped lollipops with white stems, each lollipop a different, bright, saturated color . . .



It's good!

But . . . Nah! Looks fake.

Click to expand...

But it's sharper! :-D

Rob de Loe said:

aloli said:

Rob de Loe said:

...

Click to expand...

Click to expand...

Photography has included composites since the very beginning. I've always thought of these as art pieces made with photographs.

Click to expand...

I think that's a reasonable way to think about it! If I apply the same standards to my own images though, I begin to hesitate: I've used non-generative fill, clone stamping, and similar tools in the past.

Maybe the conclusion is that those modified images are no longer photographs? Would it matter how much of the image is untouched? Perhaps dodging and burning, even in a darkroom, could be considered a composite? When physical object is used to obstruct light when printing, does the presence of the object and its role in the formation of the image preclude the end result from being a photograph?

I don't pose these as rhetorical questions, I'm again not too sure about what my answers to these would be. It would be reasonable to me, even if my eventual answers lead me to conclude that many of my images can not be considered photographs.

Rob de Loe said:

Kelli Connell's Double Life series continues to amaze me. https://www.kelliconnell.com/new-page-5 The end product is a composite made with photographs, rather than a photograph, but that sits well with me because it's a human making rather than output from a generative AI tool. Sadly, her body of work in this project is large enough now that I suspect it would not be hard to train a generative AI tool to make new images that look like they belong in the series.

Click to expand...

Thank you for sharing this - it is an interesting example because the manipulation is central to the theme of the series. It does remind me that the choice of medium can contribute to the end product in various ways, whether it serves as a tool to enable the artists' creative vision, or whether the medium itself becomes more than a means to an end.

Rob de Loe said:

I wrote a bit about the implications of generative AI for photography a couple years ago. https://www.robdeloephotography.com/Pages/Generative-AI

I know that some people see a brave new world of image making using generative AI. I see the end of most kinds of photography that I value.

Click to expand...

Very well written. In your recommendations on how to continue practicing relevant photography, I find some of my own concerns.

As you wrote, there remain some types of images and scenes that generative AI struggles with. I've worried about whether some of the more abstract and surrealist images I produce still have value, when the location is unidentifiable and a viewer would not be able to distinguish it (AI artefacts aside) from a generated image. All the viewer would have is my word that the scene depicted truly occurred, and that I photographed it myself.

aloli said:

Rob de Loe said:

aloli said:

Rob de Loe said:

...

Click to expand...

Click to expand...

Photography has included composites since the very beginning. I've always thought of these as art pieces made with photographs.

Click to expand...

I think that's a reasonable way to think about it! If I apply the same standards to my own images though, I begin to hesitate: I've used non-generative fill, clone stamping, and similar tools in the past.

Maybe the conclusion is that those modified images are no longer photographs? Would it matter how much of the image is untouched? Perhaps dodging and burning, even in a darkroom, could be considered a composite? When physical object is used to obstruct light when printing, does the presence of the object and its role in the formation of the image preclude the end result from being a photograph?

Click to expand...

I think we're in the same mind space here in that we're both just trying to figure out our own guardrails.

I'm not interested in imposing my guardrails on other people; everyone can do what they want. Rather, I find it helpful to be clear with myself about what I'm doing.

aloli said:

I don't pose these as rhetorical questions, I'm again not too sure about what my answers to these would be. It would be reasonable to me, even if my eventual answers lead me to conclude that many of my images can not be considered photographs.

aloli said:

Kelli Connell's Double Life series continues to amaze me. https://www.kelliconnell.com/new-page-5 The end product is a composite made with photographs, rather than a photograph, but that sits well with me because it's a human making rather than output from a generative AI tool. Sadly, her body of work in this project is large enough now that I suspect it would not be hard to train a generative AI tool to make new images that look like they belong in the series.

Click to expand...

Thank you for sharing this - it is an interesting example because the manipulation is central to the theme of the series. It does remind me that the choice of medium can contribute to the end product in various ways, whether it serves as a tool to enable the artists' creative vision, or whether the medium itself becomes more than a means to an end.

aloli said:

I wrote a bit about the implications of generative AI for photography a couple years ago. https://www.robdeloephotography.com/Pages/Generative-AI

I know that some people see a brave new world of image making using generative AI. I see the end of most kinds of photography that I value.

Click to expand...

Very well written. In your recommendations on how to continue practicing relevant photography, I find some of my own concerns.

Click to expand...

Thank you. I don't know how your mind works, but I find that my best shot at achieving clarity on something is to write about it.

aloli said:

As you wrote, there remain some types of images and scenes that generative AI struggles with. I've worried about whether some of the more abstract and surrealist images I produce still have value, when the location is unidentifiable and a viewer would not be able to distinguish it (AI artefacts aside) from a generated image. All the viewer would have is my word that the scene depicted truly occurred, and that I photographed it myself.

Click to expand...

We do seem to be in violent agreement. Here's another article that gets at the exact issue: https://www.robdeloephotography.com/Pages/Trust

Fundamentally, all we have is our reputations. Your images have value to you, and people who know you and your ethics will trust that what you are showing exists in the world.

I face a very similar challenge. I especially enjoy creating photographs like this one. I've shown it on this forum before and the consensus view is I should flip it around. But it's meant to be like this. It looks like a double-exposure, but it's a single frame that takes advantage of tilt to get a nearly pan-focal image. The viewer must take my word that this is real and not generated, because my word is all I've got.


Vernal Pool, Crawford Lake Conservation Area, southern Ontario

JimKasson said:

Rich42 said:

xtm said:

Wouldn't it be easier to just ask AI to generate a bucket of lollipops?

Click to expand...

I asked ChatGPT to

"Create an image of a shiny metal pail of multi-colored, cellophane-wrapped lollipops with white stems."



It misunderstood the context of "multicolored." or I miscommunicated the idea. It's a great image!

But it's not the bucket of pops in the restaurant for which the brochure is being made.

Now if I could only actually find such lollipops as these and have the restaurant use them instead . . . !

Click to expand...

The right hand stick is missing its lollipop.

Click to expand...

It's upside down in the pail!



--
Rich
"That's like, just your opinion, man." ;-)