AI and Photography: When Does an AI Image Stop Being a Photograph?
When does a photograph stop being a photograph?
It is a question that has become much harder to answer since generative artificial intelligence entered the mainstream.
Digital photographers have been having arguments about manipulation for decades. We have debated Photoshop, compositing, cloning, retouching, HDR, focus stacking and all manner of techniques that would once have been considered unacceptable.
Generative AI has changed the argument because it can create parts of an image — or an entire image — that were never in front of a camera at all.
So where should we draw the line?
The image that sparked a debate
In 2023, German artist Boris Eldagsen entered an AI-generated image into the Sony World Photography Awards.
The image, PSEUDOMNESIA: The Electrician, went on to win the Creative category, but Eldagsen declined the prize after revealing that the image had been generated using AI.
He had deliberately entered the image to provoke a discussion about the future of photography and artificial intelligence.
It certainly worked.
The story received extensive international media coverage and prompted a much wider discussion about whether an AI-generated image should be competing alongside photographs.
The interesting part wasn’t simply that AI could produce a convincing image.
It was that the image had been entered into a photography competition.
That distinction matters.
What actually makes a photograph?
For most of the history of photography, there has been a relatively simple relationship between the subject and the finished image.
Light from the subject reaches a camera sensor or film.
The photographer decides where to stand, what lens to use, how to compose the image and how to expose it. The resulting file or negative can then be processed and edited.
Of course, the final photograph may bear little resemblance to the original capture.
A photographer can crop it, adjust the colour, remove distractions, combine exposures or make substantial changes in Photoshop.
But there is still an underlying photographic capture.
Generative AI introduces another possibility: pixels can be created without the camera ever seeing the subject.
That is where the argument becomes much more complicated.
Three very different types of image
It helps to separate three things that are often lumped together as “AI photography”.
1. A photograph with traditional editing
A camera captures the image and the photographer subsequently edits it.
This might involve:
- colour correction
- exposure adjustments
- cropping
- retouching
- cloning
- dodging and burning
- focus stacking
- HDR
- compositing
There is still a photographic capture at the heart of the image.
2. A photograph with generative editing
The photographer starts with a real photograph but uses generative AI to alter part of it.
For example, an AI tool might:
- remove a person from the background
- extend the edges of an image
- replace a background
- add an object
- remove an unwanted object
- alter part of the environment
The finished image contains pixels that were not captured by the camera, but the photograph itself provided the starting point.
This is where individual competition rules become particularly important.
3. A wholly AI-generated image
Here there is no photographic capture at all.
The creator might type a description into an AI image generator and produce an image of a person, landscape, animal or imaginary scene without taking a photograph.
It might look exactly like a photograph.
It might even be more technically perfect than many real photographs.
But there was no camera involved.
For a photography competition, that difference is fundamental.
What about Photoshop?
This is where things become less straightforward.
Photographers have always used tools that create pixels that weren’t present in the original capture.
Photoshop’s traditional tools can reconstruct missing areas, remove objects or generate patterns and textures.
More recently, generative features have made that process dramatically more powerful.
Adobe Firefly, for example, can generate content from text prompts and can be integrated into photographic workflows.
So is using Generative Fill fundamentally different from cloning part of the sky?
There isn’t a universal answer.
It depends on what the image is intended to represent and, importantly, what rules apply to it.
A commercial advertising image may have completely different requirements from a documentary photograph.
A creative portrait competition may have different rules again.
And a qualification panel may have its own definition of acceptable manipulation.
The competition question
This is probably where the debate matters most to photographers.
A competition isn’t necessarily judging whether an image is attractive.
It may be judging photographic skill, creativity, technical ability, storytelling or some combination of these.
If the competition is specifically designed to assess photography, there needs to be some agreement about what constitutes a photograph.
That doesn’t necessarily mean that every competition should ban AI.
It means that the rules need to be clear.
Some competitions may welcome AI-generated artwork.
Others may allow AI-assisted editing but require the original photograph to remain the basis of the entry.
Others may prohibit generative AI completely.
The important thing for photographers is to read the rules rather than make assumptions.
The rules are becoming clearer
The Sony World Photography Awards provides a good example of how competition rules have developed since the Eldagsen controversy.
Its current rules allow manipulation, but require the extent of manipulation to be described. They also state that entries must originally be from an analogue or digital photograph and that computer-generated content cannot be the origin of the entry.
In other words, significant digital manipulation can be allowed while a completely computer-generated image is excluded.
That provides a useful distinction between editing a photograph and creating an image from nothing.
Other competitions may draw the line somewhere else.
Does AI make photography less creative?
Not necessarily.
A photographer can use AI as another creative tool without asking it to do all the creative work.
For example, a photographer might take a portrait, use AI to experiment with possible backgrounds and then use the result as inspiration for a real shoot.
An architectural photographer might use AI during the planning stage to show a client how a finished campaign could look.
A commercial photographer might use generative tools to extend a background to fit a particular advertising format.
None of those uses necessarily diminishes the photographer’s creativity.
The important question is what the tool is being used to do.
The grey areas will keep growing
The difficult cases are the ones between the obvious extremes.
Consider a real photograph with an AI-generated sky.
What about an AI-generated reflection?
What if AI removes a distracting object?
What if it adds an object that was never there?
What if the photographer uses their own photograph as the reference for an AI-generated scene?
What if only a tiny part of the final image was generated?
And what if an AI feature becomes so deeply integrated into Photoshop that the photographer doesn’t even think of it as “AI” anymore?
These aren’t theoretical problems.
The technology is developing faster than many existing definitions of photographic manipulation.
That means competition organisers, qualifications bodies and photographers themselves need to keep revisiting the issue.
There is also a question of honesty
Whatever your personal view on AI, there is a strong argument for being transparent about how an image was created.
If a client believes they are buying a photograph of something that actually existed in front of a camera, that is a different proposition from commissioning an entirely generated image.
Likewise, if a competition asks for a photograph and specifically prohibits computer-generated content, submitting an AI-generated image without declaring it isn’t simply another creative interpretation of the rules.
It is a failure to follow the rules.
Adobe has also introduced Content Credentials as part of its approach to AI transparency. Qualifying Firefly-generated content can carry information indicating that generative AI was involved in its creation.
That won’t resolve every question about AI, but provenance and transparency are likely to become increasingly important.
AI doesn’t make the debate go away
There is a tendency to divide the argument into two camps: those who believe AI is going to destroy photography and those who believe it is simply the next stage of digital technology.
Neither position is particularly helpful.
AI is a tool.
Like every other tool, its value depends on how it is used.
The more important question for photographers is what they are claiming an image represents.
If the image is presented as an imaginative piece of digital artwork, there is considerable creative freedom.
If it is presented as a photograph of something that happened in the real world, the distinction becomes much more important.
And if it is entered into a competition that is intended to assess photographic ability, the rules have to determine where the boundary lies.
So, when does an AI image stop being a photograph?
There probably isn’t one answer that will satisfy everyone.
For some purposes, a photograph can quite reasonably include substantial digital manipulation.
For others, the underlying pixels must have come from a camera.
And an entirely AI-generated image may be better described as AI-generated artwork rather than photography, however convincing it looks.
The technology will continue to make the boundaries harder to see.
That makes clear rules, honest descriptions and a good understanding of how an image was created more important than ever.
The debate that began with photographers arguing about film versus digital, or Photoshop versus traditional darkroom techniques, has simply moved on to its next stage.
The difference this time is that the question isn’t just how much can we change a photograph?
It is:
How much of the photograph needs to have been photographed in the first place?
