AI-generated content 101 — In photography
It was the year 2014 when Ian Goodfellow pitted two separate Neural Networks against each other. And thus were invented GANs.
Now, all manner of industries is scrambling to make sure they are in a prime position to take advantage of AI-generated content.
For the photography industry — from stock images through to high-end fashion spreads — there is a heady mix of excitement about the potential time, and therefore money, savings that AI could offer; and a frisson about the creative potential.
The stock image industry is projected to reach US$6.4 Trillion by 2027. AI is already playing a role in helping editors get the right image match at Getty Images. However, it is also aiding stock library owners in reducing their costs by replacing the manual processing of images.
As deep-learning platform Pimloc explains, this includes “validating / editing metadata from photographers; checking image content and quality; reformatting; retouching; cropping; tagging and captioning”.
Using AI also results in more uniformity.
For organizations with massive, and inaccessible archives, AI has allowed the digitization and categorization of thousands of images (or hours of video footage), which they can then sell.
VideoFashion went to an AI-driven asset management solution, GrayMeta, to digitize its archive of more than 18,000 hours of fashion footage and catwalk coverage captured since 1976. Amazon Rekognition was used for image and video analysis. Feed Magazine reports: “…the library footage monetized at up to $135 per second of content”.
Stock libraries though are also playing a part in training AI. Japanese stock photo platform Pixta, reported Asahi.com, began offering AI training data in 2018.
In high demand were packages of “1000 facial photos of Japanese wearing masks”, which companies needed to train facial recognition programs.
But AI is not just organizing but making the content that image libraries sell, and companies are having to rethink the legalities of their models to facilitate this. In March, Getty introduced what it described as “an industry first Enhanced Model Release form.”
Paul Reinitz, Director of Advocacy and Legal Operations Counsel at Getty Images explained: “We must recognize that the increased use of biometric data contained in imagery to train AI/ML applications requires the need to ensure that we have obtained the model’s permission to use their image and data in this manner….”
The IP rights on the “raw material” used to train AI must be protected, argues Getty. This is, after all, how so many jobbing models and photographers make their money.
In the fashion industry, speed to market is essential, especially for high street or mass-produced brands. The time it takes to find, book, and then photograph models in hundreds of outfits is a huge expense (however fast they can change clothes).
In the pandemic, it proved impossible so a change that may have taken years happened in months. Some brands accelerated their influencer portfolio — sending clothes to social media figures and models who could shoot the pics themselves at home.
But others merged the digital and the physical. Companies like Zeekit and Vue.ai digitally dress images of real models that they hold on their database with clothes that have been photographed on mannequins or on just one model. They can alter images so that the clothes fit perfectly to the model’s size and stance; as well as lighting realistically. Zeekit has also worked with Bloomindales so customers can see a virtual version of themselves “trying on” outfits.
Rosebud.ai takes images of mannequins and turns them into models, either wholly generated by AI or digital replicas of a human model. The Pandemic opened companies’ eyes to the possibilities of this technology, and now why would they return to costlier and more time-consuming past ways?
Vogue Business reported that Kyoto-based Datagrind “…worked with Japanese golf brand Honma Golf to turn roughly 50 pictures of real models into tens of thousands of pictures of different models wearing different clothing.” As the Rosebud team boast, companies like itself can offer “content creation at the speed of thought”.
But this goes beyond the realm of fast fashion. The March 2020 cover photoshoot for Vogue Italia featured five models who had never been seen before. They were created by celebrated fashion photographers Mert and Marcus. Ida, Teresa, Helga, Anastasia, and Stevie joined the ranks of the burgeoning number of virtual fashion models, including Shudu who headed up a Balmain campaign. For this campaign, there was no photographer but a digital creator.
So should photographers be packing up their Hasselblads? Fashion photographer and digital experimenter, Nick Knight, has always opened his eyes to possibilities and suggests no.
AI can be used to simplify and “mechanize” some of the less appealing aspects of the business, but can also offer a chance to have some fun. His collaboration with German digital and conceptual artist Mario Klingemann saw images of US model Sara Grace Wallerstedt taken by Knight and then fed into an AI program.
The project prompted a discussion on whether AI can match human creativity. Perhaps though, this is the wrong question. AI doesn’t need to be in competition with our creative prowess, nor set to undermine it, but can enhance it.
Photography has developed as a medium over the centuries. Now the possibilities are evolving daily.
Join the Every AnyOne’s Community
Katie Scott is a Journalist with more than twenty years of experience writing on everything from tech startups to travel. Former News Editor of Wired.co.uk, now based in Sussex after a spell in Hong Kong.
