ReWeave

Inspiration

If you ever stand at the back of a clothing factory and watch the cutting tables for a few minutes, the thing that surprises you the most is not the noise or the speed. It is the bins. Every time a worker lays a pattern over a roll and cuts out the shapes that become a onesie or a t-shirt, the space between those shapes falls away into a bin as offcuts. Depending on the garment, somewhere between 15 and 30 percent of every roll ends up in that bin. It is clean, brand-new fabric. It has never been worn, never been washed, and in most cases, it is one known material with no buttons, no zippers, and no seams. That is the single best type of cloth a recycler could ask for.

And almost all of it is thrown away or burned.

The reason is boring and infuriating at the same time. A bin of mixed scraps is worthless to a buyer until someone separates it by color, by fabric, rough size, and then manually creates an auctionhouse listing for it. Doing that by hand is slow, and labor is expensive, so the math almost always says send it to the landfill. Across the industry, the numbers are staggering. The fashion sector produces around 92 million tonnes of textile waste a year and is on track for roughly 148 million tonnes by 2030 (UNEP). Around 15 percent of fabric is lost on the cutting floor alone (Earth.org), and across the full yarn, fabric, and garment process, up to 47 percent of material can be wasted (Reverse Resources). Fashion is responsible for about 10 percent of global carbon emissions and a fifth of global wastewater, and the industry loses around 100 billion dollars of raw material every year from scraps. A truckload of textiles is landfilled or incinerated every second.

We built Reweave because the problem is not a lack of demand for this scrap. The problem is that nobody has a fast, cheap way to sort it and then sell it. So we made the sorting and listing creation instant. One framing point shaped the whole project. A brand like Carter's does not run its own factories. So Reweave is not a tool a brand uses on its own floor. It is something a brand rolls out across its supplier factories as part of a sustainability program, the same way brands already set environmental standards for the suppliers they work with. The factories use it day to day, and the brand is the one that requires it. That’s how we fit the Make & Remake problem statement of eliminating waste before it is created, and connecting one manufacturer's leftovers to another's raw material. Companies like Carters make money from this because factories turn the excess scrap into income, and then that income generates more money for the factory, which allows Carters to decrease the prices of their own products, as factories would sell clothing to the distributor for less.

What it does

Reweave is a web platform that turns a factory's fabric scraps into a marketplace. It is not just a place to list fabric. It is a scan-to-inventory pipeline. Its core job is to transform one photo of an unsorted pile into structured, searchable, priced, geolocated inventory that recyclers and makers can buy. Here’s how it works. A worker on the factory floor logs in, points a phone or a fixed camera at the scrap pile, and takes a single photo. They crop it to the scraps plus a small calibration marker, and that is the only thing they ever do. From that one photo, Reweave outlines every separate piece, reads each piece's color, estimates its fabric type and weight, and groups everything into clean lots by color family, material family, and pattern. Each lot gets a price and a number for the carbon and water saved by reusing it instead of making fabric from scratch. By the time the worker sees the words "Materials listed," real inventory already exists and the fabrics are being sold on the marketplace.

From there, the platform splits across three roles. Recyclers and makers browse a buyer marketplace with search, fabric and color filters, a grid or map view, and a quantity slider that lets them reserve an exact amount in grams or kilograms, then claim it and check out. For factory administrators, they have their own page that shows revenue, live inventory value, margin, fabric mix, run history, and a full impact report, with the evidence chain running straight from the original scan image to the lots it produced to the carbon and water those lots represent. There is also a built-in assistant that can explain any page or the recycling logic behind it.

The short version is a messy pile of fabric goes in, and clean, labeled, priced, claimable lots automatically come out in a marketplace, with the factory worker doing nothing but taking a picture.

How we built it

The architecture is a React and Vite frontend, a FastAPI backend in Python, and a Supabase Postgres database reached through SQLAlchemy. The computer vision work is coded in Python because that is where the computer vision libraries are.

The heart of this project is the computer vision pipeline. When a photo comes in, we decode it, resize it, and find a scale reference, using an ArUco, so we can convert pixels into real centimeters and estimate the size of the fabric scraps effectively. We then send the image to Segment Anything (SAM and SAM2) running on Replicate, which outlines every distinct piece without any training. Raw segmentation is messy, so a big chunk of our effort went into cleaning the masks: we reject tiny specks, background-sized blobs, the table surface, the calibration marker, and duplicate masks of the same piece. For each surviving piece, we extract a bounding box, contour, area in real square centimeters, dominant color, secondary color clusters, a pattern label, a shape label, and a cropped thumbnail.

Fabric type is the hard part, so we treated it carefully. We send the full image plus a numbered contact sheet of the pieces to an OpenRouter vision model and force it into a strict JSON schema with broad, honest categories like cotton jersey, denim, fleece, and polyester knit, rather than pretending we can read an exact fiber blend from a photo. While in the past fabric estimation has been difficult using AI models, recent research shows that LLM’s can predict fabric type with around an 80% accuracy. We then estimate weight from visual area, fabric GSM, and a fold factor, group pieces into sort bins using a consensus step so that small classification wobbles do not fragment the pile into too many bins, and render an annotated sorting plan.

We also added a physical signal. A small Arduino sits beside the table with an onboard infrared and color sensor, and on every capture, it reads how much infrared bounces back off the pile, along with the color and brightness up close. Glossier synthetics reflect more infrared, and matte naturals reflect less, which gives us a real-world material cue to sit alongside the camera's guess. What makes Reweave different

There are good companies in this space, and looking at them is what convinced us the gap was real. Deadstock marketplaces like Queen of Raw and Nona Source move whole unused rolls and excess inventory that is already sitting, labeled, in a warehouse. They do important work, and Queen of Raw alone is addressing an estimated 120 billion dollars of idle textiles. But the model assumes you already know what each item is, and it asks the seller to photograph every style and set a price by hand, often plugging into a brand's inventory management system to do it. They handle the easy, known, full rolls. They do not touch the messy mixed-scrap bin on the cutting floor, which is where most of the volume and most of the friction actually is.

On the other side are the hand-sorting operations. FABSCRAP, which we admire, trucks pre-consumer scrap to a Brooklyn warehouse where volunteers sort it by fiber and color by hand, because, in their own words, there is no mechanized system for sorting textiles by fiber yet. They sort roughly 2,000 pounds a week that way. It works, but it is slow, location-bound, and capped by how many people you can get to a sorting table.

So everyone either handles fabric that is already sorted and labeled, or relies on humans to do the sorting and the data entry. Reweave automates the exact step that everyone else leaves to people. The demand on the other end is real, and it rewards exactly what we produce. Pre-consumer cutting scrap is the premium feedstock for recycled cotton because it is clean, uncontaminated, and easy to sort, and Textile Exchange notes that the largest volume of recycled cotton already comes from pre-consumer sources. Recyclers pay more for clean, well-segregated waste, the recycled cotton market was worth about 5.8 billion dollars in 2025 and is growing at roughly 9 percent a year.