AI-Assisted Photography

[smoke665]

But they scrape images of other photographers (from social media, photography sites, almost any source they can get) and they effectively use them to learn "this is what a young woman with blonde hair and big breasts who is nude and smiling looks like

I know you do a lot of portraits, so you're saying you were born with all your photography knowledge (exposure, lighting, composition, posing, editing, etc.), never studied (scraped) any of that knowledge from others? What sets AI apart from task oriented algorithms is its ability to mimic the human process of learning, its doing exactly what we've done since man drew the first stick figures on a cave wall, but you seem to want to hold it to a different standard.

it would not be a photo taken by you.

This seems to be a sticking point with many. When it comes to inputting code its hard for the artist crowd to understand that creativity is not in the medium but in the artist. Currently AI is limited, it can't do anything without input from a human. There's been some progress in becoming self aware, (capable of formulating rational thought or understanding emotion), but its a slippery slope. To much independent thinking and you're faced with the "Singularity Scenario" - where AI achieves recursive self-improvement and operates independently of human oversight, eventually replacing us entirely.

 

[JoeW8]

I know you do a lot of portraits, so you're saying you were born with all your photography knowledge (exposure, lighting, composition, posing, editing, etc.), never studied (scraped) any of that knowledge from others? What sets AI apart from task oriented algorithms is its ability to mimic the human process of learning, its doing exactly what we've done since man drew the first stick figures on a cave wall, but you seem to want to hold it to a different standard.


This seems to be a sticking point with many. When it comes to inputting code its hard for the artist crowd to understand that creativity is not in the medium but in the artist. Currently AI is limited, it can't do anything without input from a human. There's been some progress in becoming self aware, (capable of formulating rational thought or understanding emotion), but its a slippery slope. To much independent thinking and you're faced with the "Singularity Scenario" - where AI achieves recursive self-improvement and operates independently of human oversight, eventually replacing us entirely.

First, all forms of "AI" don't "learn" the same way humans do (at least at present). For instance, a database where you ask it to do a review of the literature or a consensus of what the literature says--that's a pure case of GIGI (garbage in, garbage out), it's not learning. It's simply selecting from the data you've put in it. If you put good and comprehensive data in to it, then the recommendations are likely to be accurate. If you didn't, you get garbage. No learning by the "AI" instead it's sorting through the data sets.

Likewise, some other types of "AI" simply follow rules. Like using a program to review your writing and spot spelling or syntax errors (and you tell it to use 'American' english rather than 'English' english). Again, it's following the rules you set. Or it's analyzing the complexity of the language (and reporting that you write at a 7th grade level or whatever).

What you're referring to is generative "AI" and your example doesn't apply. I wasn't born with all my photographic knowledge--of course not. But I don't necessarily learn the same insights that you do (or anyone else) by looking at the work of others. For instance, most photographers look at a colorful sunset and go "I must take a picture" and I go "ho-hum, it's not worth taking a picture of unless the sunset is only a backdrop for something else" (like a couple kissing, or a beached boat in the foreground or someone fishing on a pier with the sunset behind them). Generative AI does NOT learn the same way that we typically learn. Some of what we learn is indeed by repetition. But generative AI "averages" or blends what it sees. That's why "good" generative AI doesn't have 1 or 5 or 10 examples to use. What are considered to be good generative AI programs have thousands of examples. And it goes through those and picks out an average or blend that fits the prompts that you specify. That's why some generative AI art programs create humans with a thumb and 5 fingers. Or an extra leg. It isn't going "this is a great example, I'm going to mimic it." It goes "there are 500 pictures in my file of a dog running so I'll look at the prompts and shape an image that meets those prompts for a dog running on a beach." And you end up sometimes with a dog that has two tails. Or five legs.

I don't deny that generative AI programs can produce really creative images. And I don't deny that it's a real skill to generate prompts that produce an image that matches what you intended. But it's not photography in my book. And it's not creativity as a photographer generating photos, it's creativity as computer technician generating a visual image that you saved as a jpeg.

 

[acparsons74 11]

Who will you be teaching? Professionals, amateurs, children, college students, etc? How will you be teaching? Person to person in a classroom, privately, videos on the web, Zoom, etc.

I'll be teaching university students in a classroom. The plan is to teach photography(composition etc...) for the first half of the term and then get into AI for editing, and finally have students generate fake artificial work that looks like a photograph.

 

[DanOstergren38 3]

@smoke665 Accusing someone of believing they were born with their photography knowledge and skill seems a little aggressive, don't you think? Just because they disagree with your stance on AI?

 

[smoke665]

First, all forms of "AI" don't "learn" the same way humans do (at least at present). For instance, a database where you ask it to do a review of the literature or a consensus of what the literature says--that's a pure case of GIGI (garbage in, garbage out), it's not learning. It's simply selecting from the data you've put in it. If you put good and comprehensive data in to it, then the recommendations are likely to be accurate. If you didn't, you get garbage. No learning by the "AI" instead it's sorting through the data sets.

Granted, but we were specifically discussing AI text to image where the database is random in nature and the algorithm assembles it's own dataset. One of my bigger concerns regarding AI generated text, is the ease with which it can be prejudice/bias tainted by the input.

What you're referring to is generative "AI" and your example doesn't apply. I wasn't born with all my photographic knowledge--of course not. But I don't necessarily learn the same insights that you do (or anyone else) by looking at the work of others. For instance, most photographers look at a colorful sunset and go "I must take a picture" and I go "ho-hum, it's not worth taking a picture of unless the sunset is only a backdrop for something else" (like a couple kissing, or a beached boat in the foreground or someone fishing on a pier with the sunset behind them).

You just confirmed the learning process. Here's a good example from back in 2024.
I didn't have a restaurant with all the right features and color, nor could I find anything in stock photos. The same photo could be created with your colorful sunset...if I remember the colors and sky, I can regenerate it with text to image as a background for a couple shot in studio. Whether you learn something immediately or it spawns an idea later it's still learning.

generative AI "averages" or blends what it sees.

My understanding is that it's learning, not just from the data set, but also from the one giving the instructions. I've found the more I use it the more closely it matches what "I want" the first time.

And it's not creativity as a photographer generating photos, it's creativity as computer technician generating a visual image

Never said it was, I said it's the artist who ultimately provides the creativity. You wouldn't try and differentiate a painter as not having real creativity because they painted something from the mind (unless it was really bad, LOL), you wouldn't differentiate a sculpture as not having real creativity because they worked with clay, you wouldn't differentiate a master carver as not having real creativity because they worked with wood or stone. In the case of text to image, the correct term is AI Artist, calling them less is demeaning.

I'm very cautious as regards AI, I saw all the Terminator movies, and we're only 3 years away from 2029. For me the potential for copyright infringement is nothing compared to the challenges moving forward. Already mentioned is the prejudice or bias possibility. Another is the truth in advertising issue. If you generate an image of the ideal woman based on the preference of your target group, and then used that image in an advertisement for your product, without a clear understanding to the viewer that it's AI generated, that's deceptive advertising. As @acparsons will be teaching the next generation of AI users, I hope that he includes a little on the challenges, as well as just how to use it.

 

[smoke665]

@smoke665 Accusing someone of believing they were born with their photography knowledge and skill seems a little aggressive, don't you think? Just because they disagree with your stance on AI?

Having seen much of JoeW's work over the years, I can say he's very talented. My point was not to be aggressive, but to point out that we all learn from others at some point, and that AI mimics that process. Just debating the other side of the coin.

 

[JoeW8]

Having seen much of JoeW's work over the years, I can say he's very talented. My point was not to be aggressive, but to point out that we all learn from others at some point, and that AI mimics that process. Just debating the other side of the coin.

I didn't regard any of your comments as aggressive. So no issue there.

And I think we miss the point when we just talk about general "creativity." I can be creative with generative AI on a computer but those are not photographic skills and they do not generate what I would consider photographic art. It's visual art. It may be incredibly creative. But I didn't generate it with my camera and I relied on the AI (using text prompts) to dramatically change what I may have submitted.

 

[smoke665]

I didn't regard any of your comments as aggressive. So no issue there.

Thank you. I think a healthy debate helps everyone understand something thats going to be with us for the future.

I think we miss the point when we just talk about general "creativity." I can be creative with generative AI on a computer but those are not photographic skills and they do not generate what I would consider photographic are. It's visual art. It may be incredibly creative. But I didn't generate it with my camera

So if i understand your thoughts correctly, its not creativity, but labeling an AI image as photography? I can agree. Maube moving forward there needs to be some new definitive AI created imaginary terminology, such that photography doesn't get marginalized.

 

[cgw133 158]

Thank you. I think a healthy debate helps everyone understand something thats going to be with us for the future.


So if i understand your thoughts correctly, its not creativity, but labeling an AI image as photography? I can agree. Maube moving forward there needs to be some new definitive AI created imaginary terminology, such that photography doesn't get decimated.

Party's over for some areas. Editorial shooter friends report AI-doctored files aren't selling well. Either it's so subtle or totally absent but I'm not seeing it gaining ground in the market where the $$$ is. Recent NYT fashion supplements aren't awash in AI images. That's top-shelf work.
Amateur/hobbyist stuff is another matter.

 

[Rickbb81 20]

I use a Canon 5D MKIII from 2012. I would love to know how AI assists my photography through this camera (asking genuinely from curiosity).

The software that renders the sensor's pixels and transforms them into a file, jpg, raw, etc. can be called AI. Whether you set the parameters manually, (creating an instruction set), or let the in-camera software do it on auto you are using an early type of what is called AI today.

You create pre-programmed functions on the camera, in doing that you are basically creating a rudimentary AI instruction set. When I prompt you, (press the shutter, while under this function), then apply that set of instructions to the pixel data set and print the result to the screen.

I realize I'm being confusing by comparing the in-camera software to AI, but really, they are the same thing. Just some programmed queries and instructions against a database. The database of pixels on a sensor or 1,000-acre data farm, same-same, just different scale.

Which is why, IMHO, that what is passing as AI, is in fact not intelligence at all. Just lots of preprogrammed instructions waiting for someone to string them together into a query. It only works the way it does is the scale, untold amounts of data all being stored, sorted, filed, resorted, repeat ad-nauseum at the speed of electricity, enormous amounts of electricity.

Anyway, (sorry one beer too many into this), back to the OP's post. How do we use AI assist? If you shoot digital, you already are in some fashion, or another was my point.

 

[AlanKlein98 14]

Not quite right. While AI uses existing images to learn patterns, styles, composition, etc., it does not typically "copy and paste" them. Instead, it creates an entirely new images based on the probabilistic patterns it learned from the training data. Using your violin example, if I tell AI to create Vivaldi's Four Seasons, then yes you would be correct, but if I told AI to compose "a violin concerto to celebrate the passing of the seasons, with emphasis on changing tempos, and a more modern blending of traditional techniques with electronic, pop, and rock styles", then it would be just as original, as if a young violinist after hearing the works of numerous composers, set his own composition to music on the violin. In photography are the poses you use or the lighting setups, strictly your own inspirations or did you compile them from ideas you gleaned from someone else? When tone mapping an image, did you just come up with a color scheme all on your own, or was it the result of viewing what others have done? Humans have been "copying" the work of others (if not exactly then as a basis for another idea) long before AI came on the scene.

AI is still in it's infancy, and as such there are still bridges to cross, especially as regards protections under the copyright laws.

The US courts have ruled that AI created work cannot be copyrighted as a human has to create it to be copyrighted. It's not clear to me what happens if AI is used to edit, though.

 

[smoke665]

The US courts have ruled that AI created work cannot be copyrighted as a human has to create it to be copyrighted. It's not clear to me what happens if AI is used to edit, though.

I think it reads you cant copyright an image created entirely by AI, but you can copyright an image incorporating AI created imagery if you can prove sufficient human authorship such as editing, composition, modification, etc. There's a lot of catching up required on the legal side. Most of the AI platforms have legal restrictions on use, and it's generally considered prudent to keep AI imagery as decorative rather than those which closely resemble another's style, a real person, or brand.

software that renders the sensor's pixels and transforms them into a file, jpg, raw, etc. can be called AI.

I have to disagree with this. The algorithms in early cameras were a set of instructions, which when activated followed a strict routine wher A+B=C. AI programming uses both user input and learned data gleaned from both external data, and user preferences to formulated a result whereA+B= C or D or E or F. They are nowhere similar.

 

[TJMcG129 12]

Not quite right. While AI uses existing images to learn patterns, styles, composition, etc., it does not typically "copy and paste" them. Instead, it creates an entirely new images based on the probabilistic patterns it learned from the training data. Using your violin example, if I tell AI to create Vivaldi's Four Seasons, then yes you would be correct, but if I told AI to compose "a violin concerto to celebrate the passing of the seasons, with emphasis on changing tempos, and a more modern blending of traditional techniques with electronic, pop, and rock styles", then it would be just as original, as if a young violinist after hearing the works of numerous composers, set his own composition to music on the violin. In photography are the poses you use or the lighting setups, strictly your own inspirations or did you compile them from ideas you gleaned from someone else? When tone mapping an image, did you just come up with a color scheme all on your own, or was it the result of viewing what others have done? Humans have been "copying" the work of others (if not exactly then as a basis for another idea) long before AI came on the scene.

AI is still in it's infancy, and as such there are still bridges to cross, especially as regards protections under the copyright laws.

Unfortunately, you will never convince some people that these AI programs are not "stealing" their photos. Some honestly think that Nano Banana actually looks at all the art on the Internet, and takes a pixel or two from each one, and assembles an image that would not exist without their own stunning sunset picture on Flickr.

Ignorance is rife; it is ever thus...

 

[smoke665]

Unfortunately, you will never convince some people that these AI programs are not "stealing" their photos. Some honestly think that Nano Banana actually looks at all the art on the Internet, and takes a pixel or two from each one, and assembles an image that would not exist without their own stunning sunset picture on Flickr.

Ignorance is rife; it is ever thus...

This is why I think debate of the subject is important. To me it seems that one of the big stumbling blocks out there is the terminology used....is it photography or something else where AI is concerned. Much like the debates that ensued when digital surpassed film.

 

[TJMcG129 12]

I didn't regard any of your comments as aggressive. So no issue there.

And I think we miss the point when we just talk about general "creativity." I can be creative with generative AI on a computer but those are not photographic skills and they do not generate what I would consider photographic are. It's visual art. It may be incredibly creative. But I didn't generate it with my camera and I relied on the AI (using text prompts) to dramatically change what I may have submitted.

I wish I knew why this cannot be universally understood. There are, and have long been, people who can draw a "photorealistic" image so good that you cannot immediately tell it is a drawing. Bert Monroy's Photoshop pixel art is legendary, and when people see his work, they think it is great photography, but there is no photography involved. He builds a scene pixel by pixel using only Photoshop. I shudder when I see someone say that prompt generative AI is "just another tool for the photographer." Another tool for a commercial artist, perhaps...