It’s a fanciful little one-piece in shimmering green and aquamarine with bold fuschia shoulder accents — perfect for a night out on the town. Is this a new dress from a Milan or Tokyo collection? Nope, it was designed by an AI-powered machine, and produced by a couple of MIT graduates.
Pinar Yanardag and Emily Salvador recently launched their fashion brand Coven.ai, aiming to reimagine dress design with the help of an artificial intelligence technique called Generative Adversarial Networks (GANs).
Pattern drafting is the first and most important step in dressmaking. Designers typically start with a general sketch on paper; add styles, elements and colors; revise and refine everything; and finally deliver their design to dressmakers. Coven.ai accelerates this time-consuming and labour-intensive process by having GANs automatically generate pattern designs.
The company’s AI algorithms have so far produced two designs: the shimmering party dress and a more formal black number with pointy shoulders, an asymmetrical neck and a single bell sleeve.
The Coven.ai team first trained an AI model to randomly generate thousands of sewing patterns; then selected a few designs based on their tastes; and finally assembled fabrics and other materials to realize the design.
At the heart of the process is StyleGAN (Style-based Generative Adversarial Network), an AI algorithm NVIDIA introduced last year that has filled the internet with those highly convincing fake images of human faces, cats, landscapes and so on. StyleGAN represents an alternative generator architecture that draws insights from style transfer techniques. The algorithm can learn and separate different aspects of an image unsupervised; and enables intuitive, scale-specific control of the synthesis. To train StyleGAN so it can learn how to synthesize new dress designs, the Coven.ai team built a dataset of nearly 30,000 high-resolution dress patterns downloaded from 10 websites.
The inspiration for Coven.ai came from How To Make Almost Anything, a popular MIT course that Yanardag and Salvador attended last summer. The course provides a hands-on introduction to cutting-edge technologies — 3D printing, laser cutting, electronics, and more — for designing and fabricating smart systems.
A former postdoctoral associate at MIT Media Lab with extensive experience in machine learning, Yanardag initiated the project How To Generate Almost Anything to create novel pizza recipes, songs, dresses, perfumes, screenplays, graffiti, and even viruses — all using AI algorithms. Encouraged by the results, Yanardag and Salvador decided to focus the process on dresses, believing AI-powered fashion could have potential.
Coven.ai has thus far produced 100 affordable AI-designed dresses, and are planning to launch a kickstarter campaign for fundraising. The company also wants to enable the creation of personalized dresses based on individual customer taste. “We are working on adopting StyleGAN and adding personalization directly as a feedback loop to the algorithm so that the stuff it is generating will be automatically designed to fit your style,” says Yanardag.
The creation of Coven.ai was not intended to replace fashion designers. Yanardag says Coven.ai is considering building an AI tool that could help reduce fashion designers’ workloads, for example by automatically previewing a wide range of patterns and styles in a particular design, etc. The Coven.ai team has also collected a number of 1920s Bergdorf Goodman sketches from the Metropolitan Museum of Art’s internet archive and used these to produce regenerated dress designs using style transfer, an AI technique for recomposing images in the style of other images.
Eugena Delman, founder of New York fashion business AVA James, seconds the idea that AI can find a fit in fashion: “From a process perspective, if we could also automate the pattern or the fit process, we would love that as well. Pattern development is not cheap and it feels like getting the fit right should be automated… Utilizing AI as part of our creative process … could also generate designs that have never been done before, which is really exciting,” Delman told Synced.
Yanardag and Salvador say the main challenge they are facing is their non-fashion backgrounds. A fashion line needs more than just appealing designs to build a lasting business — effective manufacturing, retailing, marketing, and of course fundraising also play key roles. To hone these skills Yanardag has been attending fashion school on weekends and is seeking talents with fashion business experience.
The pair speculate that one reason AI has yet to be deployed in fashion may be because the tech field is highly male-dominated — something Yanardag says they’d like to change: “We want to be role models in this area, showcase these kinds of technologies and hopefully you know, inspire some other humans!”
Journalist: Tony Peng | Editor: Michael Sarazen
2018 Fortune Global 500 Public Company AI Adaptivity Report is out!
Purchase a Kindle-formatted report on Amazon.
Apply for Insight Partner Program to get a complimentary full PDF report.
Follow us on Twitter @Synced_Global for daily AI news!
We know you don’t want to miss any stories. Subscribe to our popular Synced Global AI Weekly to get weekly AI updates.