Alta Daily is using Meta's Segment Anything model to create a new digital closet experience that allows users to organize and interact with their clothing virtually.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Meta's Segment Anything Model 3 helps fashion AI app Alta Daily reduce image segmentation costs from cents per image to nearly zero while maintaining editorial-grade quality, supporting the massive volume of clothing photos uploaded daily by users worldwide.
Most people wear only about 20% of the clothes in their wardrobe, but the rest aren't useless—they're just hard to combine. Alta Daily was born from this insight. Launched in 2025, the app allows users to photograph and digitize their entire wardrobe, then receive outfit recommendations for any occasion via natural language prompts, displayed on the user's Alta avatar. The app also tracks daily wear to avoid repetition.
Founder and CEO Jenny Wang wanted the app to have a 'clean aesthetic,' displaying clothing like fashion magazine pages, which meant removing the background from every uploaded image. However, fashion images are 'one of the most complex image datasets': white sneakers against a white wall, a blue sweater on a crumpled blue blanket, poor indoor lighting—these scenarios make segmentation difficult. Additionally, the model must capture fine details of jewelry, handle reflective surfaces, and navigate around thin hangers and mannequins.
The Alta team tested multiple segmentation models across eight product categories (from sunglasses to shoes) and found that Meta's Segment Anything Model consistently delivered the best results. SAM 3, in particular, handles everything from mirror selfies to items laid flat on a carpet. Wang noted that if every image were a professional model shot, segmentation would be much easier, but the diversity of user uploads demands 'the best segmentation model.'
Cost was another key factor. Wang was 'shocked' when she initially explored external segmentation APIs that charged cents per image. For an early-stage startup with users uploading images every second, costs would quickly add up. By using SAM, Alta has processed over 20 million images without incurring high fees, allowing the team to focus resources on product experience.
The Alta team is experimenting with Meta's SAM 3D model, hoping to unlock new ways for users to interact with their digital avatars. Wang says, 'We love trying new models. We have a large fashion-specific dataset and constantly evaluate between models.' The app has gained a global user base in the US, France, Germany, Mexico, and the Netherlands.
This article's information primarily comes from Meta AI's official blog coverage of Alta Daily, a first-party source. All statements about costs, model performance, and user numbers are directly quoted from that report, without external verification.
Alta Daily's case shows that open-source foundation models like SAM can reduce costs for startups while solving complex real-world image segmentation problems, making AI-driven consumer applications feasible.
Primary report
Primary source