研究Meta AI· 06-13 · 15:33

Alta Daily 如何用 Meta 的 Segment Anything 重塑数字衣橱

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

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

[](https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/# "Go up one level")

Try Meta AI
[](https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/# "Toggle site search")

[](https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/# "Close submenu")[](https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/# "Main menu")

[BACK](https://ai.meta.com/blog/alta-daily-fashion-app-segment-anything/# "Go up one level")

Clear

Try Meta AI

Computer Vision

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

April 6, 2026•

5 minute read

Video 2

Most people wear only an estimated 20%of the clothes in their closets, but the rest isn’t useless. It's untapped potential: combinations that never quite come together because remembering what you own, let alone imagining how it all fits together, is harder than it sounds.

The AI fashion app Alta Daily was built on exactly that insight. Launched in 2025, the Alta Daily app lets users photograph and digitize their entire wardrobe. Using natural language prompts, the app pulls from a person’s digitized closet to recommend the perfect outfit for any occasion and shows them what it looks like on their personal Alta avatar. It also tracks what a person wears each day, making it easy to avoid repeating outfits.

A month’s worth of outfits on Alta’s Head of Engineering Joon Kim’s avatar, segmented by SAM 3.

At the heart of Altais Meta’s Segment Anything Model (SAM), which has been used to segment and digitize millions of outfits. From the beginning, Jenny Wang, founder and CEO of Alta Daily, knew she wanted to build an app that had a “clean aesthetic,” laying out the pieces people own as though they were on the pages of a fashion magazine. This meant removing the background from every user-uploaded image, a task that proved to be a significant technical hurdle.

“We were researching different segmentation models,” Wang says. “Fashion in particular has one of the most complex image datasets, especially because of the inconsistent nature of user-uploaded content. For example, a photo of a white sneaker against a white wall, or a blue sweater sprawled on a wrinkled blue blanket in bad indoor lighting — that’s hard to segment.”

The team also had to account for other challenging scenarios. The right model needed to capture the intricate details of jewelry and ensure any reflective surfaces didn’t alter the true color. It also needed to be able to segment around thin clothes hangers and human models.

The Alta team tested various segmentation models across eight product categories, ranging from sunglasses to shoes, and discovered that Meta Segment Anything Model consistently delivered the best results. SAM’s ability to handle a wide variety of images, from mirror selfies to items laid on a carpeted floor, has been a key factor in its success.

This shows the improvements between SAM 1 and SAM 3.

“If we knew that every image uploaded was a beautiful model shot, segmentation would be far easier, but because of the nature of user-uploaded content, we need the best possible segmentation,” Wang says. “SAM 3 enables us to create a clean, editorial-style interface that makes digital styling a seamless and enjoyable experience.”

Beyond its superior performance, SAM has also had a significant financial impact on the company. Wang remembers being “shocked” by the cost of the external segmentation APIs she first explored, which cost a few cents per image.

“That adds up incredibly quickly, especially when you think about how many images users are uploading every second,” she says. “As an early stage company, you’re focused on the best experience, but also cognizant about cost. You have to build a product people love, but know you don’t have unlimited money.”

By using SAM, the Alta team has been able to process more than 20 million images without incurring exorbitant costs, allowing them to focus on building the best possible product for their users. The app has already gained a global following, with a strong user base in the United States, France, Germany, Mexico, and the Netherlands.

In the future, the Alta team hopes new AI research can help them create an even more immersive experience. The team is already experimenting with Meta’s SAM 3D models, which could unlock new ways users can interact with their digital Alta avatars.

“We play with everything,” Wang says. “We love experimenting with new models. We have a massive fashion-specific dataset and we’re constantly running evals between models.”

With the help of AI and open source models like SAM, Alta is helping people express their personal style and make the most of their wardrobes. As Wang says, “With AI, we can finally build next-generation shopping and styling experiences.”

Download Alta Daily Read More About SAM 3

这篇还没有中文全文

该条目暂未提供中文翻译。标题/摘要已自动中译;本系统只对人工挑选的内容生成全文翻译。

挑中后 → markitdown 取正文 → 精翻 → 此处切换为译文