Skin care is getting the virtual try-on treatment. It’s one thing to virtually “apply” foundation or lipstick to find your best shade (a feature brands have been using on their e-commerce platforms for years). But it’s a whole other feat to digitally augment skin’s predicted appearance over time with the repeated use of certain serums or moisturizers.
Anastasia Georgievskaya, co-founder and CEO of Haut.AI, is a pioneering force in the use of generative AI in beauty: On Wednesday, Georgievskaya co-launched SkinGPT, a gen AI-powered platform that “simulates photorealistic renderings of finished product results and skin-care ingredient effects based on clinical claims,” she tells Fashionista. The technology can also purportedly account for factors like aging, sun exposure and air pollutants. It’s currently being offered to brands.
“The core thing about SkinGPT is it’s able to simulate high-resolution images that have a scientific backing,” Georgievskaya says. That means that, during research & development (R&D), brands are able to “upload the clinical trial data for any type of skin-care product or ingredient, and get a simulation of how this will affect or transform skin based on [said] data.”
“For skin-care brands, it’s an opportunity to showcase the expected effect of their products and give these numbers a visual representation,” she continues.
Photo: Courtesy of Haut.AI
SkinGPT currently operates as a B2B platform, working alongside brands’ internal R&D teams since they’re the closest to clinical trials and data. Before collaborating, the Haut.AI team first asks brands what their goal is: Is it to showcase product effectiveness? Bring awareness to environmental impact? Simulate the aging process? “The most frequent use case we have right now is simulating effects of their finished products based on their clinical claims,” Georgievskaya shares. Early brand partners include Ulta Beauty, Clarins, Beiersdorf and Unilever.
The technology offers efficiency, too: Once brands upload their data to the gen AI-powered system and receive the custom simulation, they can compare it to additional simulations in the future, and make product tweaks based on their findings.
Haut.AI also allows interested brands to implement the software into a consumer-facing app so that shoppers can visualize brands’ clinical claims on product efficacy for themselves. As Georgievskaya puts it, “We know about try-ons for lipsticks or foundations, but have you ever tried on a cream online?” (She declined to disclose pricing, but hints it comes at a premium cost.)
Haut.AI also offers its own consumer-friendly version of SkinGPT. Enter: Generative Skin, a publicly accessible website listing the effects of a library of skin-care ingredients. It also allows users to upload their own photos, “try on” different ingredients and get a customized understanding of how they’ll potentially react on their skin.
“With Generative Skin, our idea is to fill the gap in the education about what products are supposed to do,” she explains. “Quite often we read about very vague product results, like ‘This product helps you improve your redness,’ but to which magnitude? Or sometimes you can read, ‘There was a 10% increase in this improvement in this perimeter,’ but it’s […] quite hard to actually envision what this clinical measure is and how it translates into the observable effect.”
Photo: Courtesy of Haut.AI
Broadly speaking, many existing gen AI beauty tools (Perfect Corp., Banuba, Google Lens, FaceCake Marketing Technologies) insufficiently test with diverse skin tones and types, developing algorithms and simulations based on narrow data sets, Georgievskaya claims. That imbalance was a key driver of Haut.AI’s launch and its early interest in experimenting with gen AI.
“Historically, a lot of the data has been collected only for a defined age group, usually between 20 to 60 years old and primarily very light skin tones,” Georgievskaya explains. “You can develop algorithms using this software, but it’s not functionally relevant because it’s not capable of analyzing all skin types.”
SkinGPT and Generative Skin’s algorithm has a database of around 3 million images — including whole face pictures, selfies, micro images and hi-res lab photos — that account for extensive skin tones, types and textures. Haut.AI’s competitors also fall short, says Georgievskaya, because they implement filters across all photos, creating “inaccurate simulations.” SkinGPT and Generative Skin aim to be more precise.
“What’s different with SkinGPT is it’s a smart system: Before making the simulation, SkinGPT will analyze the baseline level of skin conditions and only after that will it perform the simulation. That’s why the simulations are photorealistic and they are more accurate,” she explains.
Photos: Courtesy of Haut.AI
For brands, the true test of SkinGPT’s effectiveness will be sale conversions. There’s not enough data to report just yet, but the platform offers templates to brand partners who integrate it into consumer-facing programs. It tracks the software’s key consumer interaction metrics, including how much time shoppers spend with the simulations, how many simulations they try, the conversion rate of adding products into baskets post-simulation, the average basket size and the average order value.
“Brands will be able to see if [SkinGPT] is driving any sales on their e-commerce platforms or if it’s driving engagement because it helps consumers understand their skin,” Georgievskaya says.
SkinGPT touts the potential waste-reduction benefits of its tech, too: While Georgievskaya doesn’t claim it will solve the beauty industry’s sustainability issues (after all, AI is one of the leading users of energy), it may help consumers disengage from overconsumption, she asserts.
According to some estimates, of the reportedly 120 billion pieces of beauty products produced yearly, more than 90% of the packaging could ultimately end up in landfills. Georgievskaya believes that, with SkinGPT’s technology helping brands better articulate a product’s function and expected results, shoppers will make more informed purchasing decisions.
“Even though generative AI can drive higher electricity consumption, it’ll help tackle other parts of sustainability,” she says. “I also think as the demand for gen AI grows […] field developers will create solutions to address the challenges in the field and help reduce its electricity consumption.”
Photo: Courtesy of Haut.AI
Looking ahead, Georgievskaya understands some brands may have concerns around employing gen AI into their software: She predicts brands’ biggest worry is consumer backlash in the event that shoppers see a simulation, buy and use the product, and don’t experience the expected results. In that case, she implores labels to liken the situation to using pharmaceutical drugs: Every individual responds differently to a product, some benefitting more than others. SkinGPT and Generative Skin project a scenario that could happen, but can’t promise it will be realized. She also emphasizes that results can take time, typically “four to eight weeks,” she says.
Educating consumers on skin-care ingredients and product use will continue to be priorities for SkinGPT and Generative Skin. Georgievskaya plans to build out the latter’s ingredient library, including listing additional ingredients — both traditional and newly discovered — and their effects (positive and negative). She wants Generative Skin to be a “one-stop shop” for knowing which ingredients optimize results and which to avoid combining. The hope is to inspire consumers to focus more on ingredient effectiveness and less on marketing when shopping for skin care.
“Tools like SkinGPT and Generative Skin empower consumers,” she says. “I think that push for function-first is something the industry needs.”
Source: Fashionista.com