AI Images in Photography Navigating Ethics and Use Cases

Many photographers react strongly to the idea of using AI-generated images in their work, often with a firm "Never!" While there are certainly many situations where using AI imagery in photography is unacceptable, let's first explore some instances where it might be okay. Afterwards, we'll look at the applications that raise more serious concerns.

Acceptable AI Image Uses in Photography

I can think of three scenarios where I believe it's appropriate to incorporate AI-generated elements into photography.

1. Printing Canvas Wraps

I once needed to create several canvas wrap prints for a client from one of my photos. She requested a specific dimension, but when I uploaded the image to the online printing service, it indicated that some of my photograph would be cropped. One option was to extend the edges with black, making the sides of the frame black, but the client preferred otherwise. What was the solution?

To address this, I used Adobe Photoshop's Generative Expand feature. This tool generated borders that blended seamlessly with the photo, appearing as a natural extension. These generated parts would only be visible on the sides of the canvas frame, not on the main image surface. Interestingly, the client later decided to have them framed, making these AI-extended edges even less noticeable.

2. Eliminating Distracting Elements From Photographic Art

Photography is often a form of art. Not all photographs are intended to be photojournalistic or historical documents. If I'm creating an art piece and there's a distracting tree branch or streetlight in the shot, I'm going to remove it.

Historically, I've used cropping, the clone tool, Content Aware Fill, or the Remove Tool. These usually work well. However, sometimes a tool like Generative Fill performs better, leaving fewer visible artifacts. Essentially, my photo has then used a small amount of AI to analyze the image and remove the distraction. Is this unethical? Is it more unethical than using Content Aware Fill? Or more so than a portrait photographer removing a blemish from someone's nose?

Ideally, I would have composed the shot to avoid the distracting element in the first place. But we all know this isn't always physically possible.

I believe most of us can agree that if we are creating photos for photojournalistic or historical purposes, nothing should be removed.

3. Fixing Perspective

The rationale for using AI-generated borders here is conceptually similar to the canvas wrap example: we are expanding the borders to achieve a technical correction, not to deceive.

Occasionally, when photographing buildings, I can't avoid keystoning (where parallel lines appear to converge or diverge). Sometimes, I also can't physically move back far enough to shoot wider, which would give me room to correct the keystoning in post-processing without losing too much of the image edges.

A simple solution for this is to expand the edges using Generative Expand. While Content Aware Fill can be used, it rarely looks as good as Generative Expand and is often more time-consuming to get a great result.

I'll expand the edges first, then correct the keystoning. It's fine if I lose the AI-generated edges during this process. In fact, most or all of the AI-generated content usually disappears, though some might remain in the corners. This approach is often much easier and yields better-looking results than Content Aware Fill or other methods.

As mentioned, the best approach is always to shoot as wide as possible. But in real-world scenarios, this isn't always feasible. It's okay to give yourself some grace and use the technology available (again, for photographic art, not for photojournalism or historical photographs).

When AI Image Use Becomes Problematic

If you find the above uses questionable, you'll likely strongly object to "deep fakes" and other instances of blatant deception. Here are a few applications which, in my opinion, are never acceptable.

Deception

It's clear that some individuals and groups use AI-generated images to deceive. This has significantly eroded trust. For instance, some people no longer believe my night sky photos are real. Far more seriously, misinformation and deception campaigns have damaged public trust on a wider scale. If people can no longer believe what they see or read, what are the consequences for society?

Theft

Photographers and other artists often argue that because AI models are trained on existing work, they are essentially mimicking or stealing other people's creations without permission. Adobe attempts to address this by training its models only on Adobe Stock images and public domain content without copyright restrictions. Some argue that humans also learn and synthesize from existing art, making AI's process similar. However, this argument doesn't sit well with many. Being inspired by others' art is one thing; ingesting and regurgitating it without purpose or genuine feeling is quite another.

Loss of Revenue and Job Security

There's considerable debate about the extent of revenue loss AI-generated images might cause. It's unlikely to have benefited the microstock industry.

Energy Consumption

A compelling case can also be made against using AI due to its incredible consumption of electricity and water. The International Energy Agency (IEA) projects that electricity demand from data centers globally is set to double by 2030, reaching 945 terawatt-hours. This amount is slightly more than the entire current electricity consumption of Japan.

For this reason, I have noticeably reduced my use of ChatGPT, AI-generated images, and similar tools. However, because AI is now implemented in so many services, it's increasingly challenging to avoid using it, even when making a phone call to the dentist or performing an online search.

Finding the Ethical Balance with AI in Photography

A strong guideline is this: if you believe any reasonable viewer of your photograph would feel deceived if you told them how you created it, then you've likely crossed an ethical line.

Beyond that, it's a matter of personal comfort. I feel comfortable using AI-generated images in my photography occasionally for the specific purposes described above. However, I respect that many would not. Since recently learning about AI's significant energy consumption, I will be extremely mindful of how I use it.