AI-generated photos and art have started to overrun stock image websites, social media, and other mediums, as we had previously documented in our essay on the future of photography. This trend is expected to continue in the near future. Some of them have even been awarded medals for their artistic merit, and the general trend is only moving in the direction of more application.
Now, in what appears to be an effort to combat at least some of this flood to their image-selling platforms, a few of the major stock photo websites that, up until relatively recently, allowed artificial intelligence-generated imagery to be sold on their platforms have begun applying ban hammers and other restrictions.
This had already begun in the past with more specialized and quality-focused picture-selling websites like PurplePort, whose proprietors swiftly implemented their own restrictions and explained their reasons in a blog post about the topic.
However, at that time, sites such as Getty and Shutterstock, amongst others, continued to permit this kind of image content without restrictions. In point of fact, up until quite recently, a search conducted on Shutterstock alone using the keyword “AI-generated” got well over 18,000 hits. Also, keep in mind that this only included photographs that were freely described by their uploaders as being generated using artificial intelligence (AI).
Since then, this has shifted, and it appears that it will continue to develop to a considerable degree. For its part, Getty Images, the largest stock picture repository in the world, has determined that it will no longer accept new submissions generated by AI-image generators and that it will delete those that are already present on the site. This decision was made effective immediately. Petapixel received a copy of an email that the agency had written to all of its contributors informing them of the new AI image prohibition, and the agency shared this email with the photo site.
We recommend checking out our post on our own research in AI-rendered picture synthesis for a peek at the outcomes, which range from eerily bizarre to completely nonsensical. This will provide a little bit of background information on this topic. Additionally, we have investigated what this type of technology portends for the foreseeable future of photography.
These kinds of pictures are not examples of digital artwork that were pieced together by human artists using their hands. Instead, as our post explains in detail, they are automatically created graphical representations of straightforward text prompts that users can submit to artificial intelligence art-rendering websites such as Mid journey, DALL-E, and Stable Diffusion, amongst others. These websites allow users to submit their creations to be rendered.
It is reasonable to assume that Getty and other companies will take great care to differentiate between digital artwork that was produced almost immediately by AIs through their own algorithmic procedures and digital artwork that was created by humans utilizing software tools.
Getty claims that it is being careful to make the distinction in its own statement, which reads as follows: “These changes do not prevent the submission of 3D renders and do not impact the use of digital editing tools (such as Photoshop, Illustrator, etc.) with respect to modifying and creating imagery.” [T]hese changes do not prevent the submission of 3D renders.
In addition to this, the photographs that are produced by these AI systems are not yet protected by copyright in the United States. This is one of the reasons that Getty cites in its message to content providers as a consideration in their decision: “There are unanswered problems with regard to the copyright of outputs from these models, and there are unmet rights issues with respect to the underlying images and metadata utilized to train these models,” says one researcher.
There are real concerns with respect to the copyright of outputs from these models and unaddressed rights issues with respect to the imagery, the image metadata, and those individuals contained within the imagery, Getty CEO Craig Peters also explained in a statement that was nearly identical to the one given to The Verge.
These statements, coming as they do from a company that has a long history of engaging in truly dirty games with regard to how it applies copyright to images from creators, illegally claiming copyright over imagery that is in the public domain, or engaging in many other pretzel-like twists with copyright law, are interesting, to say the least.
In any event, because Getty has implemented its prohibition, the AI-photo pool that was previously available on the site has practically vanished. As at the time this article was written, the number of results returned by a search for the phrase “mid journey” (which refers to one of the most well-known AI rendering systems) had decreased from thousands to almost nothing.
It appears that other prominent stock picture websites are following suit and jumping on the bandwagon. Although Shutterstock has not yet implemented a blanket ban on AI-generated imagery, it does appear to be restricting searches for specific categories of AI-created content.
It is important to take note that Getty’s platform still contains certain examples of artificial intelligence (AI) art. There may have been a significant drop in the appearance of certain results, while others are still present. Both “AI-generated art” and “mid journey” were terms that none of our searches turned up any results for. A search for a shorter keyword, “AI art,” did, however, give well over a thousand hits; nonetheless, it is likely that the majority of these results are examples of digital art generated by humans.
The issue is, however, that it is not always easy to determine, and the difficulty of making that determination is only likely to increase in the future.
The CEO of Getty argues that the company is dependent on its customers to recognize AI-rendered art and report it to the website. Additionally, the business plans to collaborate with other organizations, such as the Coalition for Content Provenance and Authenticity, in order to develop new filters that are capable of detecting these kinds of inventions in advance.
For the time being, the majority of AI-rendered imagery exhibits visible distortions in ways that made its source difficult to conceal. The results of our own trials with it establish this point quite clearly. The results of works of visual art that do not strive to replicate photographs in an obvious way might be more ambiguous and difficult to differentiate from the products of human creativity.