How Alta Daily Uses Meta’s Segment Anything to Reimagine the Digital Closet

People wear a fraction of what's actually hanging in their closets, and it's not because those clothes are junk. It's because putting outfits together from memory is genuinely hard. That's the problem Alta Daily set out to fix when it launched in 2025: an app that lets you photograph your entire wardrobe, then leans on natural language prompts to recommend outfits and preview them on a personal avatar. It also logs what you've worn so you're not repeating the same combo every week.

The technical backbone of all this is Meta's Segment Anything Model, which founder and CEO Jenny Wang says has been used to segment and digitize millions of outfits. Wang wanted Alta to feel like flipping through a fashion magazine, which meant stripping backgrounds cleanly from every photo a user uploads. That's a harder problem than it sounds. A white sneaker photographed against a white wall, or a blue sweater tangled in a wrinkled blue blanket under bad lighting, is exactly the kind of chaotic input that breaks segmentation models. Add jewelry with fine detail, reflective surfaces that can distort color, thin clothes hangers, and human models in the frame, and the difficulty compounds fast.

Wang's team tested multiple segmentation models across eight product categories, everything from sunglasses to shoes, and found Meta's SAM held up best across the messiest, most inconsistent user photos, whether that's a mirror selfie or a sweater dumped on a carpet. She's blunt about why this matters: if every upload were a polished studio shot, any decent model would do. But real users don't shoot like that, so the bar for segmentation has to be higher. SAM 3, she says, is what lets Alta keep its editorial-style look without users doing any extra work.

There's also a cost story here that Wang doesn't gloss over. Early on, she explored external segmentation APIs and was shocked at prices running a few cents per image, a number that balloons fast once you're dealing with constant user uploads. For an early-stage company trying to build something people love while still watching the bank account, that math matters. Using SAM instead let Alta process more than 20 million images without the runaway costs, freeing the team to focus on product rather than budget triage. The app has since picked up users across the United States, France, Germany, Mexico, and the Netherlands.

Wang's team isn't stopping at 2D segmentation, either. They're already experimenting with Meta's SAM 3D models to see what a more immersive avatar experience could look like, running constant evaluations against their own fashion-specific dataset. Wang frames it simply: AI is what finally makes next-generation shopping and styling experiences possible, and for a wardrobe app built on solving a very ordinary problem, that ambition tracks.