Cheryl
Liu

Fifty colorways before lunch, and not one of them stitched.

A designer at one of Raspberry AI's customers, a man named Edvin, can sit down in the morning and test fifty fabrics or colorways by the afternoon. None of them get physically sampled. None of them cross an ocean in a FedEx box. The whole exercise that used to eat weeks and a sample budget now happens on a screen, and the outputs are not mood-board approximations. They are photo-realistic renders that look the way the garment would look on the brand's own website.

That is the product Cheryl Liu sells, and the easiest way to understand why it matters is to watch a general-purpose image generator fail at it. Type "fuzzy sweater" into Midjourney and you get something sweater-shaped and vaguely soft. Type it into Raspberry AI and the system knows the knit, the gauge, the pile, the dozen pieces of vocabulary a working designer never has to explain to another designer. "There's a lot of terminology behind that sweater," Liu likes to point out, "that a Midjourney does not know." Fashion, it turns out, is a language, and she built the model that speaks it.

Raspberry AI is an end-to-end generative-AI platform for fashion product development. Designers feed it sketches and trend research; it returns photo-realistic product images, 2D technical drawings, and CAD files that a manufacturing team can actually build from. It will spin up synthetic customer groups to survey a design before a stitch is committed. The point is not to make pretty pictures. The point is to collapse the distance between an idea and a thing a brand can sell.

To understand why that distance matters, look at the workflow Raspberry AI is replacing. Designers and merchandisers juggle a stack of disconnected tools and time-consuming manual steps: a hand-drawn sketch, then a round of stakeholder reviews, then tech-pack generation, then a physical sample, then shipping, then another review. The bottleneck is structural, and it leaves brands chronically a step behind the trend cycle - behind the influencer-driven appetite for frequent, fresh product drops. Raspberry AI's bet is that compressing the iteration loop does more than save money. It changes what a brand is capable of noticing and answering in the market.

She moved the week the tools arrived.

DALL-E and Stable Diffusion landed in late 2022. Most of the fashion industry was still arguing about whether AI belonged anywhere near the design room. Liu was already building. She had spent her early career as a private-equity analyst at KKR, where the focus was retail and she put money into large fashion enterprises - the kind of companies that run thousands of stores and live or die on how fast they can read a trend. Then came product and machine-learning roles at Amazon's AWS, at DoorDash, and at Catawiki. Finance taught her the economics of a slow design cycle. Big-tech taught her how models actually get built and shipped.

When the image models became good enough, she had the rare combination to act: she knew, from the inside, exactly how much money and time a sample room burns, and she knew what it took to fine-tune a model that would not embarrass a designer. She founded Raspberry AI and pointed all of it at one bottleneck - the iterative, manual, tool-fragmented grind between a hand-drawn sketch and a tech pack.

She has a BA from Columbia and an MBA from Stanford, but the credential that matters most to her customers is that she invested in fashion before she sold software to it. She understood the buyer because she had been adjacent to the buyer's balance sheet.

The machine is the brush, not the painter.

It would be easy to read a fashion-design AI as a replacement for designers, and Liu spends real energy arguing the opposite. Her framing is that Raspberry AI gives designers superpowers - a way to multiply creativity and productivity without sacrificing originality. The drudgery is what gets automated: the sampling, the shipping, the endless back-and-forth between creative, technical, merchandising, and vendor teams. What stays human is the part that was always human.

"Human intuition, cultural context, and artistic vision remain irreplaceable," she says, and she means it as a product principle, not a platitude. The most successful design teams she sees are the ones that fold AI in thoughtfully - using it to delete the tedious middle and keep the creative core at the center. Andreessen Horowitz, which led the Series A, bought the bigger version of that idea: that generative AI can be fashion's great equalizer, giving a small label the iteration speed that used to belong only to giants.

From seed to Series A in about ten months.

Raspberry AI raised a $4.5M seed round, then roughly ten months later closed $24M in Series A led by a16z, with Greycroft, Correlation Ventures, and MVP Ventures along for the round. Total reported funding runs north of $28.5M. The customer roster reads like a department-store directory: Under Armour, J.Crew, Tapestry (Coach, Michael Kors, Kate Spade, Stuart Weitzman, MCM), TJX, lululemon, MCM Worldwide, the Italian manufacturer Gruppo Teddy with its thousands of stores, and the supply-chain giant Li & Fung.

The recognition followed the revenue. Raspberry AI was named to CB Insights' 2026 AI 100, selected from more than forty thousand companies and standing as the only creative-AI application on the list. It was picked as one of a handful of AI developer finalists for the inaugural CFDA x OpenAI Innovation Hub. Liu herself was named a 100 Women in AI honoree. And in a move that turned the software into theater, Raspberry AI debuted a first-of-its-kind AI-driven collaboration with the designer Theophilio at New York Fashion Week, SS26 - real clothes on a real runway, built through a process that did not exist a few years ago.

What a16z saw, and what it wrote down when it led the round, was a founder with a deep technical approach, encyclopedic fashion-market knowledge, and a commitment to building one of the most comprehensive AI design platforms anywhere. The firm had concluded that horizontal models would always lack the domain-specific detail that apparel demands, and that the iterative design and prototyping phases were where AI could land the heaviest blow. Inside customers like Li & Fung and Under Armour, that has translated into measurable outcomes - faster design cycles, more SKUs explored, quicker market entry, and less overstock sitting in warehouses. The platform became, in a16z's word, integral.

Apparel was the wedge, not the ceiling.

Liu has said Raspberry AI plans to push past clothing into home, furniture, and cosmetics product design. The thesis travels: any category where a creative team iterates, samples, and ships physical product is a category where the concept-to-market cycle can be compressed. The secret sauce is not an off-the-shelf wrapper but proprietary in-house models and fine-tuned ControlNets tuned to produce outputs a retail team can manufacture from.

It is a strange and specific kind of ambition: not to make fashion look more like AI, but to make AI fluent enough in fashion that nobody on the design floor has to think about it. The tell is in how she talks about her own customers. The win is not that the model is clever. The win is that Edvin tested fifty colorways and went home on time.

Four things worth knowing.