ZMO.ai - The End of Apparel Models?
Fashion Tech
We’re back. Wifi has been rough here and have found myself doing everything else but writing.
We’ve spoken about the world of fast fashion and how it is only getting faster. SHEIN (NextRound article here) is leading the way with hot and “in” pieces at crazy low prices. As an aside, recently they have received consumer backlash for their environmental transgressions in a WSJ article here (s/o SG and HD). As a society, we are so focused on speed, convenience, and efficiency. Amazon Prime has two-day shipping, we can download games and movies in seconds/minutes, and have much of the world at our fingertips. Today we talk about deepfake modeling tec
For the fashion space, one of the biggest bottlenecks happens to be models.
Clothing companies need to find models that embody their brand’s image to display their clothing. They need to find talented photographers to shoot. And they need to find the space for these shoots to happen. And this happens thousands of times, for companies around the world, for each new season, for each new line, each new piece (SKU), each and every year. With the power of technology, machine learning, and computer vision, this too is a space that is being innovated. And while it may save clothing companies time and money, it may hurt the models that make a living by inspiring others to buy the clothes!
ZMO.ai, a Chinese-founded startup raised an $8MM Series A round on May 17, 2022 led by GL Ventures, with participation from GGV Capital and GSR Ventures to generate high-quality, AI models for clothing companies that can be customized with the touch of a few buttons.
Of course, ZMO.ai appeals to small e-commerce businesses that can not hire or afford models for every piece that they create, but dang, ZMO.ai is likely to disrupt the entire modeling space. With an infinite stream of new styles, changing seasons, and new clothing concepts, would every fashion brand not benefit from a tool that creates fully detailed models of your clothing without the costs, pain, and overhead of using real models, photographers, and spaces? I mean damn.
According to TechCrunch,
On average, e-commerce companies spend around 3%-5% of their annual gross merchandise value (a rough metric measuring sales, usually excluding returns and refunds) on photoshoots, according to Roger Yin, who worked at Evernote and ran his own cross-border e-commerce business before co-founding ZMO.
The tech behind ZMO.ai is based on generative adversarial networks (GANs), the same secret sauce behind controversial deepfakes and the world of misinformation. I studied deepfakes at Duke in my computer science curriculum, and the tech is truly mindboggling.
A quick few sentences from the college paper:
Deepfakes are a form of synthetic media that utilize complex computer science concepts to depict something that never *actually* happened. It enables the creation of a digital copy of something, someone, etc. Deepfakes can appear in photos, videos, audio clips, and far more from sophisticated algorithms known as general adversarial networks (GANs). The concept of a general adversarial network was devised by Ian Goodfellow in 2014, essentially giving computers the ability to create things from scratch as opposed to “recognizing” things as they do in time and resource-intensive unsupervised learning (ML). In a general adversarial network, two neural networks compete against one another to create fake images and determine which images are real and fake, much like a forger and a detective would. The first net, termed the “generator” produces artificial outputs such as photos and videos that are as realistic as possible. The second termed the “discriminator”, compares these generator products with those from the original dataset and tries to determine which are real and which are fake. The generator tunes its parameters to create more and more realistic outputs, the discriminator improves its requirements to become more discriminating and the two nets go back and forth until the discriminator can no longer determine what’s real and what’s fake. In essence, they both push each other to improve their resulting outputs.
It’s pretty incredible.
In the case of ZMO, the software enables brands to generate virtual full bodies of models by defining simple parameters like face, height, skin color, body shape and pose. All that it takes is just some decent photos of the piece you are interested in putting onto a model. In addition to time and cost savings, ZMO also enables brands to create a more inclusive brand image by visualizing pieces on ethnically diverse models. This in turn can improve the conversion of targeted marketing to your customers!
Ella Zhang, ZMO’s CEO and co-founder and a former engineer at Google and Apple says
“Traditionally, the entire cycle of garment manufacturing may take two to three months, from design, fabric selection, pattern making, modeling, to actually hitting the shelves. We are flipping and shortening that process. [Brands] can now put it on a virtual model, [and put it on] the website. Once orders come in, the e-commerce company can start manufacturing,”
What’s more, despite how long it takes to manufacture clothes and advertise them, sales cycles are extremely short in the clothing space, meaning marketing needs to be effective and have high conversion so that consumers buy your product before they buy someone else’s!
Unsurprisingly, ZMO.ai is working on a partnership with SHEIN, the fastest “fast-fashion” brand that pushes out 45,000+ new items a week! If any customer is a perfect use-case for ZMO.ai, it’s SHEIN.
The model (pun intended)
ZMO’s business model is a credit model (like what iStockPhoto uses), much like tokens at an arcade.
There are various tiers, with each pricing tier getting you two basic things:
Access to a certain number of “models” - this is the number of digital models / people you can have try-on / model your clothes
Access to a certain number of “credits” - this is the number of standard definition images you can generate using the platform
The larger “credit-pack” you purchase, the cheaper it is per-credit. This in turn encourages users (or in this case, businesses) to buy more credits.
Dark side?
ZMO.ai has created a product that will revolutionize how brands promote their items. Being based in China means there are data and privacy concerns that are of customer (and investor) concern. But the larger question is how will deepfake tech affect photographers, models, and other adjacent / related individuals?
Additionally, being able to artificially “simulate” or generate images that show your clothing on ethnically diverse individuals: does that mean that a brand is actually inclusive and promotes diversity? I don’t think that constitutes genuinity if you can just use software to create images without any human input…
I think the deepfake software behind ZMO is groundbreaking that will change more than just fashion marketing. It is going to change how we perceive the world around us - in the news, in entertainment, in social media, pretty much everything we use on the internet. I hope we can use the tech for the right reasons and although the nature of the tech renders this borderline-impossible for most consumers, there are clear ways to delineate the difference between what is real, and what is “Deepfaked”.
The solo euro journey
I’ve been traveling solo for the last two weeks through Europe. Have jumped from Mallorca, Spain to Puglia & Ischia, Italy. Have seen some gorgeous beaches and landscapes, eaten some insane food, and met some awesome people from all around the world. My luggage actually got lost from the flight for 4 days in Italy but being here has been so awesome it was hard to be in a bad mood over it.
Most importantly, I’ve had many, many negronis - and they obviously are phenomenal in Italy, the homeland of the aperitif. It’s been such an eye-opening and empowering journey traveling alone and I couldn’t recommend it more.
On one of my first days in Mallorca, I visited a beautiful beach named Calo des Moro, a hidden beach tucked away in a small cove, a hike away from publicly accessible roads. There’s a steep / rocky / slippery path you need to traverse down to get from the hike to the actual sand and people of all ages were on this beach. It was quite impressive to see.
While I was there, there was a group of girls on the beach taking photos together but they were not speaking to one another whatsoever. After watching for a few moments, I realized they were using sign language to communicate with one another. What to take photos of, where to position themselves, how the photos were coming out, and more were all being communicated via sign language.
Made me appreciate how easily and efficiently many of us can communicate and realize how we can take one of life’s greatest gifts for granted every now and then.
Today I head off to Turkey to join up with some buddies as we do Istanbul, Bodrum, and Telaviv, Israel! Dropping some more photos of my journey below:
GW
interesting, lots of 3d tech going into apparel, enhanced by ml & ai. we're working further back in the chain, as we manufacture our goods. we have used a similar tech, except it used live models who we had hire for a certain number of 'shots'. the rest of process is similar to this tech. still not quite ready for what we need but getting close. lots of folks playing in this area, who ever gets there first has a lot to gain. exciting work, great article and you look as if you've got the work/life balance figured out! enjoy, all the best. gale (your cousin from the great white north)
Fascinating article, thanks. Amazing trip, enjoy!!! Elaine