AI image generators have exploded in popularity. But how exactly do they work? And why are some people raising concerns about their usage?
Generating images with the use of artificial intelligence (AI) is something that has been widely discussed in recent years. Stories range from positive reviews of how impressively capable this new technology is to fears that it marks the beginning of the end for the photography and creative industries.
But how do they work? And what can they be used for? In this article, we take a closer look at the technology to answer these questions and explore how it has already managed to gain a bad reputation in the photography industry. But first – what exactly is an AI image generator?
What is an AI image generator?
An AI image generator, otherwise known as a text-to-image generator, is a piece of software that uses AI to create digital images from scratch using text prompts input by a human user.
AI image generators have many uses, from functional jobs such as redesigning the interior of your apartment, creative projects like producing fine art, or bespoke images for advertising to potentially more sinister uses such as creating deepfakes.
How do AI image generators work?
At the most basic level, a user inputs a number of keywords into a piece of software and a digital image based on those keywords is created.
This may sound simple, but an awful lot of technology would have been used to create and train the software.
Just like a human, a computer cannot create an image of something it has never seen. Therefore, every AI image generator has been trained on millions, if not billions, of digital images to understand what things look like. It then uses this knowledge to make an educated choice about what to draw when prompted by text keywords.
Without such training, AI image generators would not be fit for purpose. For example, if you had never seen what a cat looks like – or heard a description of one – trying to draw an accurate representation from the name alone would prove difficult. AI image generators can encounter the same problem.
For further reading, we recently wrote an article that focused on Google’s SR-3 AI-powered image upscaling technology, which uses very similar techniques.
Can anyone use AI image generators?
Yes and no. In most cases, you do not need to be a software engineer to use AI image generators. Indeed, there are already reports of AI image generator prompts for sale on marketplaces, which makes it even easier for users to create high-quality results.
The more sought-after technology, however, is often subject to waiting lists and paywalls. Anyone can register for access, though, so once the model has been properly tested, you should expect to see AI image generators readily available to the general public.
Examples of AI image generators
Three of the main AI image generator models are DALL-E, Stable Diffusion, and Midjourney. They each have different ways of working and, indeed, different results. Take a closer look below:
DALL-E
Author: OpenAI
Website: https://openai.com/blog/dall-e/
Stable Diffusion
Author: StabilityAI
Website: https://stability.ai/blog/stable-diffusion-public-release
Midjourney
Author: Midjourney
Website: https://midjourney.gitbook.io/docs/
What’s the problem with AI image generators?
Arguably the main problem with AI image generators concerns the general lack of regulation around the technology.
Perhaps the most widely discussed issue is the harm that can be caused by deepfakes, although the problems with the technology are broader than this, and could bring serious issues for the photography industry on many levels.
Deepfakes
Some generators prevent a user from creating images that include celebrities or other famous faces. But as the technology develops and becomes democratized, it is impossible to deny the threat of disinformation that can arise from this.
This article by a photographer who used AI to imagine how dead celebrities would look if they were still alive today shows the remarkable possibilities. Furthermore, the video below shows the somewhat unnerving capabilities of the technology in video.
Learn more: Deepfake videos have us concerned, but are we overlooking a more sinister threat found within them?
Copyright infringement
There are two main issues around copyright and AI image generators. The first is whether the images that are used to train the software have been licensed. The second is the issue of who owns the copyright to the final image.
Image training
As AI image generators are trained using existing digital images, there is also a question over how these images have been sourced – and whether they have been properly licensed.
For example, take a look at this article from The Verge, which includes evidence of an AI image generator reproducing the Getty Images watermark. This suggests the software is being trained using images that have not been paid for.
Furthermore, the question of whether valid model releases have been obtained raises another issue around personal data misuse.
All this evidence has prompted Getty Images and PurplePort to ban AI-generated images from their platforms – and Shutterstock is following suit.
Intellectual property
Another potential problem is the question of who owns the rights to the final image that is produced.
The US Copyright Office recently dismissed a claim by an AI image creator who tried to attribute the rights to an AI-generated image to the algorithm that created it.
US law says that works can only be protected by copyright if they were created by a human. Therefore, if a computer, a monkey, or some other non-human author was responsible for it coming into existence, it is not possible for anyone to claim ownership over that work.
This is, however, a contentious issue that we would expect to continue evolving as the technology grows.
Taking work away from creators
While the above issues focus on specific pieces of AI-generated imagery, there is a wider concern over the photography and creative industries as a whole.
As this technology matures and becomes more capable, people may no longer see the benefit of paying for creative talent.
This issue of democratizing creativity is one that could have a significant impact on not just the photography industry, but also art, CGI, architecture, and much more.
Why is the AOP concerned about changes to the UK’s copyright framework?
It is perhaps unsurprising that many potential issues surrounding AI image generators have caught the attention of the Association of Photographers (AOP), a UK-based organization that promotes and protects the rights of photographers, which has recently released a statement regarding the proposed exception to the UK’s copyright bill – the Text and Data Mining Exception.
In its own words, the AOP states that: “Currently, the Text and Data Mining exception (to copyright protection) permits non-commercial purpose machine analysis of online content, provided that there is lawful access (such as a subscription). It is also limited to prevent the resale or reuse for other purposes and must be accompanied by an acknowledgment of the source.
“This new proposed Text and Data Mining exception for commercial purposes – by the UK government – undermines this by freely allowing the machine mining of all imagery published online for any use by anyone, including AI developers. It would cover both copyright works and those protected by the UK Database Rights.”
It goes on to talk about “serious economic consequences”, saying that the proposal “completely short-circuits the licensing process allowing AI developers and others free commercial access to content for which, under normal circumstances, they would have to license and pay for.”
Identifying AI bots and crawlers as being able to scrape images from creators’ websites and social media pages in an instant, the AOP concludes that this “change in UK legislation would fundamentally turn the tables on creators giving way to economically harmful competition by allowing a content ‘free for all’ and invoking an unfair machine-endeavour vs. human endeavour scenario.”
What does it all mean for the digital imaging industry?
With so much still to be decided, it is hard to predict just how AI image generators will affect the imaging industry.
With bottom lines being squeezed tighter than ever, many commercial organizations will welcome the arrival of a cheaper alternative to traditional methods. But will the appeal of truly human creation ever die? The character that comes with an artist’s unique personality and life experience is something that is impossible to emulate.
With photography, it’s easy to imagine concerns over commercial usage in genres such as fashion or fine art. But in most cases, photography is used commercially as a way to document real-life occurrences or locations, such as a wedding day, a news event, or a travel destination – and it would be extremely difficult to effectively replace this kind of photography with an AI image generator.
There is, of course, also a certain magic in capturing the beauty of a moment you have witnessed. And when you consider this, it’s difficult to imagine a world of humans without photography.
To learn more about how SmartFrame image-streaming technology can help you protect online images while making you money through in-image advertising, get in touch with our friendly team today.
Subscribe to our newsletter
Get the latest articles and updates delivered straight to your inbox