Deep Agency shows the dangers of applying AI to the fashion industry

Deep Agency shows the dangers of applying AI to the fashion industry

Deep Agency shows the dangers of applying AI to the fashion industry

Generative AI is disrupting industries — with understandable controversy.

Earlier this month, Danny Postma, the founder of Headlime, an AI-powered marketing copy startup recently acquired by Jasper, announced Deep Agency, a platform he describes as a “photo studio and agency of AI mannequins”. Using art-generating AI, Deep Agency creates and offers “virtual models” for rent starting at $29 per month (for a limited time), allowing customers to place the models on digital backgrounds for do their photo shoots.

“What is deep agency? It’s a photo studio, with some big differences,” Postma explained in a series of tweets. “No camera. No real people. No physical location… What’s the point? Tons of stuff, like content automation for social media influencers, templates for marketers’ ads and e-commerce product photography.

Deep Agency is very much in the proof-of-concept phase, that is to say… a little stuffed. There are a lot of artifacts in the models’ faces, and the rig places guardrails — intentional or not — around which physiques can be generated. At the same time, the creation of Deep Agency models is strangely difficult to control; try to generate a female model wearing a particular outfit, like a policeman, and Deep Agency just can’t do it.

Nevertheless, reaction to the launch was quick – and mixed.

Some Twitter users applauded the technology, expressing an interest in using it to model clothing and apparel brands. Others accused Postma to pursue a “deeply unethical” business model, scraping photography and likenesses of other peoples and selling them for profit.

The divide reflects the broader debate over generative AI, which continues to attract astonishing levels of funding while raising a host of moral, ethical, and legal issues. According to PitchBook, investments in generative AI will reach $42.6 billion in 2023 and soar to $98.1 billion by 2026. But companies such as OpenAI, Midjourney, and Stability AI are currently in facing lawsuits for their generative AI technologies, which some accuse of reproducing the work of artists without compensating them fairly.

Picture credits: deep agency

Deep Agency seems to have particularly struck a chord because of the application – and implications – of its product.

Postma, who did not respond to a request for comment, is not shy that the platform could compete with — and possibly harm the livelihoods of — real-world models and photographers. While some platforms like Shutterstock have created funds to share revenue from AI-generated art with artists, Deep Agency has taken no such action – and hasn’t signaled its intention to do so. TO DO.

Coincidentally, just weeks after Deep Agency’s debut, Levi’s announced that it would be partnering with design studio to create custom AI-generated models to “increase the diversity of models that buyers can see wearing its products”. Levi’s stressed that it plans to use the synthetic models alongside human models and that this decision would not impact its hiring plans. But it raised questions as to why the brand hasn’t recruited more models with the various characteristics it seeks, given the difficulty these models have historically had in finding opportunities in the fashion industry. the fashion. (According to a survey, in 2016, 78% of models in fashion advertisements were white.)

In an email interview with TechCrunch, Os Keyes, a PhD student at the University of Washington who studies ethical AI, observed that modeling and photography — and the arts in general — are particularly vulnerable fields. to generative AI because photographers and artists lack structure. power. They are largely low-paid independent contractors for large companies looking to cut costs, Keyes notes. Models, for example, often have to pay high agency commissions (~20%) as well as business expenses, which can include airfare, group accommodations, and promotional materials needed to land jobs with clients.

“Postma’s app is – if it works – actually designed to kick the already precarious creative worker chair further and send the money to Postma instead,” Keyes said. “It’s not really a thing to applaud, but it’s also not hugely surprising… The fact is that socio-economically, tools like this are designed to dig deeper and focus the profits.”

Other critics take issue with the underlying technology. State-of-the-art image generation systems, such as those used by Deep Agency, are called “broadcast models”, which learn to create images from text prompts (e.g., “a sketch of ‘a bird perched on a windowsill’). navigate through training data retrieved from the web. The problem in artists’ minds is the tendency of broadcast models to essentially copy and paste images – including copyrighted content – from the data that was used to form them.

deep agency

Picture credits: deep agency

Companies selling broadcast models have long claimed that “fair use” protects them in the event that their systems are trained on licensed content. (Enshrined in U.S. law, the fair use doctrine permits limited use of copyrighted material without first obtaining permission from the rights holder.) But the artists allege the designs infringe their rights, in part because the training data was obtained without their authorization or consent. .

“The legality of a startup like this is not entirely clear, but what is clear is that it aims to put a lot of people out of work,” said Mike Cook, an ethicist from AI and a member of the open research group Knives and Paintbrushes. said TechCrunch in an email interview. “It’s hard to talk about the ethics of tools like this without addressing deeper issues related to economics, capitalism and business.”

There is no mechanism for artists who suspect their art was used to train Deep Agency’s model to remove that art from the training dataset. That’s worse than platforms like DeviantArt and Stability AI, which offer ways for artists to opt out of contributing art to train art-generating AI.

Deep Agency also did not say whether it would consider establishing revenue sharing for artists and others whose work helped create the platform’s model. Other providers, such as Shutterstock, are experimenting with this, relying on a combined pool to reimburse creators whose work is used to train AI art models.

Cook points to another issue: data privacy.

Deep Agency allows customers to create a “digital twin” model by uploading approximately 20 images of a person in various poses. But uploading photos to Deep Agency also adds them to training data for higher-level models on the platform, unless users explicitly delete them afterwards, as stated in the terms of service.

Deep Agency’s privacy policy doesn’t say exactly how the platform handles user-uploaded photos, in fact, or even where it stores them. And there’s apparently no way to stop rogue actors from creating a virtual twin of someone without their permission – a legitimate fear in light of non-consensual deep nude models like Stable Diffusion have been used to create.

deep agency

Picture credits: deep agency

“Their terms of service actually state that ‘you understand and acknowledge that similar or identical builds can be created by other people using their own prompts.’ That’s kind of funny to me because the premise of the product is that everyone can have bespoke AI models that are unique every time,” Cook said. photos are also passed on to others for potential use. I can’t imagine many big companies like the prospect of either of these things. »

Another problem with Deep Agency’s training data is the lack of transparency around the original set, Keyes says. That is, it’s not clear what images the model powering Deep Agency was trained on (although the confusing watermarks in its images give some clues) — which leaves open the possibility of an algorithmic bias.

A growing body of research has revealed racial, ethnic, gender and other stereotypes in image-generating AI, including the popular Stable Diffusion Model developed with support from Stability AI. This month, researchers from AI start-up Hugging Face and the University of Leipzig published a tool demonstrating that models such as OpenAI’s Stable Diffusion and DALL-E 2 tend to produce images of people who appear white and masculine, especially when asked to portray people in positions of authority. .

According to Vice’s Chloe Xiang, Deep Agency only generates images of women if you buy a paid subscription — a problematic bias from the get-go. Additionally, writes Xiang, the platform tends to create blonde white female models even if you select an image of a female of a different race or likeness from the pre-generated catalog. Changing a model’s appearance requires making additional, not-so-obvious adjustments.

“Image-generating AI is fundamentally flawed because it depends on the representativeness of the data on which the image-generating AI was trained,” Keyes said. “While it primarily includes whites, Asians, and light-skinned blacks, not all the summaries in the world will provide representation for darker-skinned people.”

Despite the glaring issues with Deep Agency, Cook doesn’t see it or similar tools disappearing any time soon. There’s just too much money in space, he says – and he’s not wrong. Beyond Deep Agency and, startups like and Surreal are getting big venture capital investments for technology that generates virtual fashion models, ethics be damned.

“The tools aren’t good enough yet, as anyone using the Deep Agency beta can see. But it’s only a matter of time,” Cook said. “Entrepreneurs and investors will keep bumping into opportunities like this until they find a way to make one of them work.”


Be the first to comment

Leave a Reply

Your email address will not be published.