The current state of AI images on stock platforms
Shortly after Midjourney, DALL-E 2 and Stable Diffusion were released in 2022, it was clear that the business model of classic stock image databases would possibly be the first to be threatened.
While the big players like Shutterstock, Adobe or Getty were still figuring out how to deal with the new technology, new interesting projects like stockai.com were launched. The operators have understood that artificial intelligence extends the range of a stock database practically to infinity. Unfortunately, after a short period of operation, they have announced that they will discontinue their service on February 25, 2023.
This is made possible by connections via API to the corresponding AI models. If no ready-generated image is available on the platform, the user can create one himself with just a few clicks.
Some platforms have also decided against supporting AI after some back and forth. The world’s largest platform of free images Unsplash, which now belongs to Getty, came to the conclusion not to accept AI-generated images. 3D renderings, on the other hand, have been okay for a while.
It is unclear how Unsplash plans to verify the creation of the images. After all, it has long been possible to use AI to imitate the style of well-known 3D programs such as Blender or Octane, so that with a bit of fine-tuning, differences are hardly noticeable.
Paid stock image platforms have also commented on AI, some positively and some negatively. Three major players have emerged: Adobe Stock, Shutterstock, and Getty Images.
But they don’t just open the doors to AI artists and let them sell their paintings there; there are important differences.
Adobe Stock
As of February 2023, Adobe Stock is the only platform that lets you monetize AI-generated images.
Of course, this is not another empty hack to supposedly make a quick fortune on the Internet. But in fact, there are the first success stories.
Kristina Kashtanova, for example, who made headlines with one of the first AI-generated graphic novels and a copyright dispute, has shared her experiences on Instagram. She has uploaded a whole mass of midjourney images. She didn’t become a millionaire in no time, but at least she can now cover the cost of her Midjourney subscription.
Certainly, their success is not only due to the mass of images. Her experience in prompt engineering also guarantees that the images show exactly what she has in mind.
Why Adobe is making this move on AI images in the first place is debatable. It is possible that Adobe, one of the leading providers of image editing programs such as Photoshop or Lightroom, is using them to train its own AI models. For some time now, many new functions in the software have been based on machine learning.
Adobe has not created a separate category for AI images, nor are they specifically labeled to make the use of the tools like Midjourney clear. Most of them are only tagged with a keyword like “ai”. However, there you can also find photos and digital graphics depicting this topic that were created purely by hand.
Shutterstock
Shutterstock is also an interesting case to look at and a prime example of how indecisive large vendors are with technology.
First they allowed artists to sell AI photos on the platform, but later reversed that move. Instead, they chose a different path, perhaps inspired by offers from StockAI.
After announcing a partnership with DALL-E 2 developer OpenAI, they announced in January 2023 that they would make their own AI generator available through their website.
While they do not go into more detail about the model used, it can be assumed that this is a finetuned version of DALL-E. The results are very similar and there can hardly be any other reason for a cooperation between Shutterstock and OpenAI.
OpenAI is not the only partner of the industry giant. Facebook’s parent company Meta also recently extended its cooperation. “By tapping into Shutterstock’s collection of millions of images, videos and music, Meta plans to use these datasets to develop, train and evaluate its machine learning capabilities,” the press release reads.
Shutterstock’s investors seem to like the development: Shortly thereafter, the stock made a gain of 12 percent within two days.
Getty Images
Finally, we want to take a look at Getty Images, because this platform is also causing a stir in the AI scene.
To make a long story short: Getty is not happy about AI. The US company has gone to court against Stability AI, the developers of the open-source Stable Diffusion model, first in the UK and a few weeks later also in the USA. Their accusation: material from Getty was used millions of times without permission to train the model, thus violating copyright. Das ist nicht die einzige Front, an der
In fact, there is solid evidence for this, not only because Stable Diffusion (like Midjourney) sometimes co-produces artifacts of Getty’s watermark.
Getty is not the only platform affected, however, as can be found out through the search engine haveibeentrained.com. This contains all images from the LAION 5-B dataset. A simple search for “Getty Images” or “Shutterstock” spits out hundreds of images with watermarks from Getty, iStock (owned by Getty as well), Alamy or Shutterstock. Adobe Stock content, on the other hand, does not appear to be included.
Why Getty is so vehement against Stability AI may not necessarily have economic reasons. According to their latest financial report from Q3 2022, they are seeing increasing numbers of paying customers, especially in their subscription offering.
The copyright debate is not just an issue for these giant companies. Individual artists, whose works were used for training without asking, have also sharply criticized the publication of AI tools from the beginning.
Conclusion
It took a few months, but by now a majority of commercial stock image providers have positioned themselves regarding AI. While Adobe Stock and Shutterstock are generally positive about the technology, Getty Images is campaigning in court against the further development of artificially generated images.
The business model of stock databases has already been the subject of fierce criticism in the photography scene for years. While some may really generate a passive side income through masses of high-quality image material at ridiculous low prices, most will probably not even reach the payout threshold.
Something similar can be expected for the development of the sale of AI images as a business idea, except that here it is even faster to produce high-quality material. A few mouse clicks and keystrokes compared to choosing a scene, taking a picture and editing it.