MarkBarbieri
Semi-retired
- Joined
- Aug 20, 2006
- Messages
- 6,166
I don't know if you guys saw the headline, but a Sony World Photography Award winner recently declined the award. It wasn’t because the creator of the photograph didn’t use a Sony camera. He didn't use a Canon or a Nikon either. In fact, he created the picture without using a camera at all. He used an artificial intelligence program to create the image.
The art of “photography” is going through a major transition. During my lifetime, I've already seen a few revolutions in photography. Some were relatively minor things that seemed revolutionary at the time, like automatic exposure and autofocus. Automatic exposure made it easy to get your exposure right and so people took fewer bad pictures, but in retrospect, that wasn’t terribly revolutionary. The same is true for autofocus, which made it easier for people to get their pictures in focus. When those features were introduced, a lot of professionals were skeptical because they could already set exposure and focus very well manually and didn’t see much advantage in doing it automatically. The main impact of the technology was to make it easier for non-professionals to take better pictures. You’ll see this theme repeated.
I think the first major revolution was digital photography – both the ability to take pictures with digital cameras and the ability to edit digital images with tools like Photoshop. At first, digital images were a step backward from film pictures in quality, but the price difference was astounding. Before I switched to a digital camera, it was costing me the equivalent of $0.75/picture in today’s dollars. It was rare that I would take more than 24 to 50 pictures in a day. With digital cameras, I quickly found myself shooting hundreds of pictures in a day.
And it wasn’t just that I could take lots and lots of pictures. Now people could edit their pictures without complex darkroom techniques. Photoshop made it relatively easy to make changes. Doing anything non-trivial in Photoshop was complicated, but it was much easier than working in a darkroom. At first, a lot of professionals were dismissive of digital editing, promoting the idea that a good photographer should “get it right in the camera.” Many professional photographers considered editing photos to be “cheating.” The main impact was for non-professional photographers. Suddenly they could take lots and lots of practice pictures at no extra cost and they could correct a lot of their mistakes with editing tools.
In a way, I think that the introduction of good cell phone cameras was also a revolution. They were a step backward in image quality, but that was made up for by the fact that almost everyone almost always had a camera almost everywhere they went. Any picture you take with a cell phone is always going to be better than a picture you didn’t take with your dedicated camera because you didn’t have it with you. And the ease of sharing pictures with phones also contributed to an explosion of picture-taking.
The result wasn’t always positive. Some classic travel spots have become inundated with people waiting in line to take a selfie to show off to their friends. Serious photographers weren’t impressed by these little cell phone cameras, but non-professionals loved them.
The new revolution is what I'll call the computational photography revolution. It started with little things. Photo editors, like Photoshop, started to automate tasks that used to be time-consuming and difficult. Adobe introduced things like "content-aware fill" that could replace sections of your picture in a way that often looked seamless. They made it simple to replace the sky in photos. They added tools to automatically smooth skin, whiten teeth, brighten eyes, and even subtly reshape faces borrowing some of the visual tricks used by makeup artists. These were things that skilled editors have been able to do for a long time, but now it is getting easier for non-professionals to make professional-looking edits.
Cell phones have gotten into the computational photography game as well. They can take amazing low-light photos with “night mode.” They have a portrait mode that blurs the background in a way that mimics the shallow depth of field of a larger camera system. They can generate smooth panoramic shots by just rotating your phone, without requiring you to load a bunch of images into your computer and blend them there. Some phones have long exposure modes that mimic the look serious photographers get with long exposure shots using neutral density filters and tripods. Some phones have a panning shot mode that makes it simple to get shots of subjects moving against their background. It’s a technique that good photographers have done for years, but it was always difficult to do well. Professionals haven’t been too excited with these tools because they don’t need them, but non-professionals love them because they can now get shots without having specialist skills.
The most extreme form of computational photography is artificially generated images. There are programs like Dall-E 2, MidJourney, and many others that let you create pictures without a camera or Photoshop editing skills. You tell the program what you want to see and then it generates an image that matches your description. With these programs, you don’t need a camera or the technical knowledge of how to use one. You don’t need Photoshop or the technical skills to edit pictures. You need to be able to describe what you want to see, and the image generator builds something that matches your description. The more creative and detailed you can describe your vision, the better your picture will be. That's how the image was created that won the Sony award. A lot of professional photographers don’t even recognize these as photographs. Many consider them to be cheating. But non-professional photographers love them because they can create images that they never could have dreamed of making before.
I’ve seen a lot of articles and videos about image-generation tools. Oddly, there is little focus on using them to edit or modify existing images. I suspect that this will become more popular over time than generating completely artificial images. Imagine taking a quick selfie and then telling your AI editing tool to change it so that it has the lighting style of Joe McNally. And replace your annoying little brother in the picture with Tom Cruise. And ask it to convert the photo into a painting in the style of Vermeer. And change it so that we’re in Paris with the Eiffel Tower in the background. I'm not sure how far we are from a version of Photoshop that replaces most of the sliders and brushes with microphone input where you just tell it what you want changed, but I don’t think it is far off. And a lot of professional photographers and editors will say that it is cheating and point out that they can get better results without it, but non-professional photographers will love it.
To be clear, the software available today is kind of amazing but also terrible. You see the occasionally amazing pictures that are generated, but when I've played with some of the tools, I struggled to get anything that didn't look awful or weird. I'm not sure if the problems will quickly be fixed or if it will be like full self-driving cars, which have been a year or two away for more than a decade.
There are also a lot of non-technical issues that need to be thought through. Today, if you use an AI tool to generate an image, you don't own the copyright for it. Nobody does. Machines can’t create copyrighted works. It's not clear to me what the boundaries are. How much can an AI tool manipulate an image that you took before you lose your copyright or how much do you have to edit and modify an AI-generated image before you can copyright it? There are also complaints that these AI tools used a lot of copyrighted images to learn how to generate images. The holders of those copyrights say that AI tools shouldn't be able to use that knowledge to generate images in the style of an artist without somehow compensating that artist. But nobody has ever been able to copyright a style, so I’m not sure how this will turn out.
There will be an interesting cultural shift as well. You already can’t trust the images you see because photo editing tools have made it possible to create realistic fake images. You used to need good editing schools to fool most people, but new tools are making it easier and easier. It won’t be long before you can Instagram your amazing vacation to Paris without having to leave home. How will this impact the value of influencers on social media when anyone can appear to visit the same locations, wear the same clothes, and drive the same cars? Maybe we can fix some of the body image issues that people face by looking at edited pictures of other people by creating smart mirrors that will digitally perform the same mutilations that people pay cosmetic surgeons to do. Mirror, mirror on the wall, make me look AMAZING! Will people be happier when they can see themselves as fake looking as the people they admire?
But beyond those intentional fakes, how will this affect collections of family photos that get passed down through generations? Today, if I look at a slide my father took in the 1950s, I’m confident that it shows what he really saw. When future generations look at those images, how will they know that my mother really looked the way she did on her honeymoon and that someone didn’t radically alter her appearance or location? I think we’ll lose not only our ability to believe newsworthy images we see today, but even our ability to trust the images that are supposedly from the past.
The change will also impact how people do business. What jobs disappear and what jobs are created? It feels like a lot of technical skills will be less valuable. On the other hand, creative skills will probably be just as valuable as ever. If you can imagine a spectacular picture but don’t have the skills to take it, you will benefit from these changes.
And how will this impact the people in front of the camera? Why should I pay for a beautiful actor when I can hire a homely one and easily make them look beautiful? And if you can replace the appearance of your actors, you can pay them a lot less because you can easily hire a replacement for them. Or maybe the experience with Milli Vanilli will apply and people will reject AI-generated actors in favor of “real” people.
Maybe the changes won’t just be to make people look better. Maybe movies will be localized for different markets by changing the appearance of actors so that they match their audience racially and ethnically. More trivially, maybe they’ll adjust people’s lip movements so that dubs don’t seem so disconcerting.
It’s hard to know what the future will look like. Will we be able to describe a movie plot and have an AI tool generate the movie for us? Will historians be able to generate realistic VR environments so that we can “experience” how places appeared in their heyday? So much is changing so quickly that it is hard to keep up with what the future will be like. I’m certainly having a hard time picturing it.
At this point, I would normally encourage you to start playing with these tools. The problem is that they’re rough right now. If you’re really interested, get a paid subscription to one of the tools. The free versions are so hobbled now that they are practically useless. But don’t expect much yet. Just be prepared that getting the pictures you want will continue to get easier and easier. And don’t throw out your old pictures if they look bad because future tools will be able to fix the problems with them. If you are doing cool things with AI imaging tools or computational photography, I’d love to hear about it.
The art of “photography” is going through a major transition. During my lifetime, I've already seen a few revolutions in photography. Some were relatively minor things that seemed revolutionary at the time, like automatic exposure and autofocus. Automatic exposure made it easy to get your exposure right and so people took fewer bad pictures, but in retrospect, that wasn’t terribly revolutionary. The same is true for autofocus, which made it easier for people to get their pictures in focus. When those features were introduced, a lot of professionals were skeptical because they could already set exposure and focus very well manually and didn’t see much advantage in doing it automatically. The main impact of the technology was to make it easier for non-professionals to take better pictures. You’ll see this theme repeated.
I think the first major revolution was digital photography – both the ability to take pictures with digital cameras and the ability to edit digital images with tools like Photoshop. At first, digital images were a step backward from film pictures in quality, but the price difference was astounding. Before I switched to a digital camera, it was costing me the equivalent of $0.75/picture in today’s dollars. It was rare that I would take more than 24 to 50 pictures in a day. With digital cameras, I quickly found myself shooting hundreds of pictures in a day.
And it wasn’t just that I could take lots and lots of pictures. Now people could edit their pictures without complex darkroom techniques. Photoshop made it relatively easy to make changes. Doing anything non-trivial in Photoshop was complicated, but it was much easier than working in a darkroom. At first, a lot of professionals were dismissive of digital editing, promoting the idea that a good photographer should “get it right in the camera.” Many professional photographers considered editing photos to be “cheating.” The main impact was for non-professional photographers. Suddenly they could take lots and lots of practice pictures at no extra cost and they could correct a lot of their mistakes with editing tools.
In a way, I think that the introduction of good cell phone cameras was also a revolution. They were a step backward in image quality, but that was made up for by the fact that almost everyone almost always had a camera almost everywhere they went. Any picture you take with a cell phone is always going to be better than a picture you didn’t take with your dedicated camera because you didn’t have it with you. And the ease of sharing pictures with phones also contributed to an explosion of picture-taking.
The result wasn’t always positive. Some classic travel spots have become inundated with people waiting in line to take a selfie to show off to their friends. Serious photographers weren’t impressed by these little cell phone cameras, but non-professionals loved them.
The new revolution is what I'll call the computational photography revolution. It started with little things. Photo editors, like Photoshop, started to automate tasks that used to be time-consuming and difficult. Adobe introduced things like "content-aware fill" that could replace sections of your picture in a way that often looked seamless. They made it simple to replace the sky in photos. They added tools to automatically smooth skin, whiten teeth, brighten eyes, and even subtly reshape faces borrowing some of the visual tricks used by makeup artists. These were things that skilled editors have been able to do for a long time, but now it is getting easier for non-professionals to make professional-looking edits.
Cell phones have gotten into the computational photography game as well. They can take amazing low-light photos with “night mode.” They have a portrait mode that blurs the background in a way that mimics the shallow depth of field of a larger camera system. They can generate smooth panoramic shots by just rotating your phone, without requiring you to load a bunch of images into your computer and blend them there. Some phones have long exposure modes that mimic the look serious photographers get with long exposure shots using neutral density filters and tripods. Some phones have a panning shot mode that makes it simple to get shots of subjects moving against their background. It’s a technique that good photographers have done for years, but it was always difficult to do well. Professionals haven’t been too excited with these tools because they don’t need them, but non-professionals love them because they can now get shots without having specialist skills.
The most extreme form of computational photography is artificially generated images. There are programs like Dall-E 2, MidJourney, and many others that let you create pictures without a camera or Photoshop editing skills. You tell the program what you want to see and then it generates an image that matches your description. With these programs, you don’t need a camera or the technical knowledge of how to use one. You don’t need Photoshop or the technical skills to edit pictures. You need to be able to describe what you want to see, and the image generator builds something that matches your description. The more creative and detailed you can describe your vision, the better your picture will be. That's how the image was created that won the Sony award. A lot of professional photographers don’t even recognize these as photographs. Many consider them to be cheating. But non-professional photographers love them because they can create images that they never could have dreamed of making before.
I’ve seen a lot of articles and videos about image-generation tools. Oddly, there is little focus on using them to edit or modify existing images. I suspect that this will become more popular over time than generating completely artificial images. Imagine taking a quick selfie and then telling your AI editing tool to change it so that it has the lighting style of Joe McNally. And replace your annoying little brother in the picture with Tom Cruise. And ask it to convert the photo into a painting in the style of Vermeer. And change it so that we’re in Paris with the Eiffel Tower in the background. I'm not sure how far we are from a version of Photoshop that replaces most of the sliders and brushes with microphone input where you just tell it what you want changed, but I don’t think it is far off. And a lot of professional photographers and editors will say that it is cheating and point out that they can get better results without it, but non-professional photographers will love it.
To be clear, the software available today is kind of amazing but also terrible. You see the occasionally amazing pictures that are generated, but when I've played with some of the tools, I struggled to get anything that didn't look awful or weird. I'm not sure if the problems will quickly be fixed or if it will be like full self-driving cars, which have been a year or two away for more than a decade.
There are also a lot of non-technical issues that need to be thought through. Today, if you use an AI tool to generate an image, you don't own the copyright for it. Nobody does. Machines can’t create copyrighted works. It's not clear to me what the boundaries are. How much can an AI tool manipulate an image that you took before you lose your copyright or how much do you have to edit and modify an AI-generated image before you can copyright it? There are also complaints that these AI tools used a lot of copyrighted images to learn how to generate images. The holders of those copyrights say that AI tools shouldn't be able to use that knowledge to generate images in the style of an artist without somehow compensating that artist. But nobody has ever been able to copyright a style, so I’m not sure how this will turn out.
There will be an interesting cultural shift as well. You already can’t trust the images you see because photo editing tools have made it possible to create realistic fake images. You used to need good editing schools to fool most people, but new tools are making it easier and easier. It won’t be long before you can Instagram your amazing vacation to Paris without having to leave home. How will this impact the value of influencers on social media when anyone can appear to visit the same locations, wear the same clothes, and drive the same cars? Maybe we can fix some of the body image issues that people face by looking at edited pictures of other people by creating smart mirrors that will digitally perform the same mutilations that people pay cosmetic surgeons to do. Mirror, mirror on the wall, make me look AMAZING! Will people be happier when they can see themselves as fake looking as the people they admire?
But beyond those intentional fakes, how will this affect collections of family photos that get passed down through generations? Today, if I look at a slide my father took in the 1950s, I’m confident that it shows what he really saw. When future generations look at those images, how will they know that my mother really looked the way she did on her honeymoon and that someone didn’t radically alter her appearance or location? I think we’ll lose not only our ability to believe newsworthy images we see today, but even our ability to trust the images that are supposedly from the past.
The change will also impact how people do business. What jobs disappear and what jobs are created? It feels like a lot of technical skills will be less valuable. On the other hand, creative skills will probably be just as valuable as ever. If you can imagine a spectacular picture but don’t have the skills to take it, you will benefit from these changes.
And how will this impact the people in front of the camera? Why should I pay for a beautiful actor when I can hire a homely one and easily make them look beautiful? And if you can replace the appearance of your actors, you can pay them a lot less because you can easily hire a replacement for them. Or maybe the experience with Milli Vanilli will apply and people will reject AI-generated actors in favor of “real” people.
Maybe the changes won’t just be to make people look better. Maybe movies will be localized for different markets by changing the appearance of actors so that they match their audience racially and ethnically. More trivially, maybe they’ll adjust people’s lip movements so that dubs don’t seem so disconcerting.
It’s hard to know what the future will look like. Will we be able to describe a movie plot and have an AI tool generate the movie for us? Will historians be able to generate realistic VR environments so that we can “experience” how places appeared in their heyday? So much is changing so quickly that it is hard to keep up with what the future will be like. I’m certainly having a hard time picturing it.
At this point, I would normally encourage you to start playing with these tools. The problem is that they’re rough right now. If you’re really interested, get a paid subscription to one of the tools. The free versions are so hobbled now that they are practically useless. But don’t expect much yet. Just be prepared that getting the pictures you want will continue to get easier and easier. And don’t throw out your old pictures if they look bad because future tools will be able to fix the problems with them. If you are doing cool things with AI imaging tools or computational photography, I’d love to hear about it.