How I AI: Yana Welinder's Workflows for an AI-Native Fashion Brand

In this episode, I was so excited to sit down with Yana Welinder, one of the few people I see on X who truly lives at the intersection of AI, art, and fashion. She’s building Yana Bana, an AI-native fashion brand where AI isn't just a tool for inspiration—it’s what she calls her “technical co-founder.” Yana is a solo founder, but she has an entire team of AI agents helping her run the business, from design and prototyping all the way to vendor outreach and e-commerce management.

What Yana is building is a look into a future where creative ideas that were once impractical or wildly expensive are now totally possible. She’s taking her unique vision and, step-by-step, using AI to bring it to life as a physical product and a real business. It’s a powerful example of how these tools can unlock creativity and entrepreneurship for anyone.

We walked through her entire full-stack workflow, and it’s one of the most comprehensive uses of AI in a single business I’ve seen yet. She covers how she turns sketches into a whole suite of fashion collateral, prototypes previously unmanufacturable garments with 3D printing, and manages the entire business backend without a human engineering team. It’s a fantastic look at what happens when you combine a creative vision with the powerful execution capabilities of AI.

Workflow 1: AI-Powered Fashion Design & Visualization

The first part of Yana’s process is all about translating a creative vision into concrete, high-quality visuals. For any designer, this is a critical step, but it can be hard to get an AI model to produce something that is both unique to your style and photorealistic. Yana’s solution is to treat her prompt like a detailed product spec.

Step 1: Crafting the “Fashion Prompt” Spec

Before generating any images, Yana developed what she calls her “fashion prompt.” This isn’t just a simple description; it's a comprehensive technical spec that defines what makes a garment successful. As a product person, I absolutely love this. It forces you to define what “good” looks like. Her spec includes details about:

• Silhouette: The overall shape of the garment.
• Proportion/Volume: The scale and space the dress occupies.
• Fabric: How it flows, behaves, and drapes.
• Construction Details: Seams, tailoring, and assembly.
• Movement & Sound: How the garment looks and even sounds when worn.

This level of detail is crucial. As my friend Zach says, “What’s good for AI is good for humans.” This is the exact information a human designer or pattern maker would need, and it’s what allows Yana to get such specific and consistent results from the model.

Step 2: Generating the Core Image

With her spec ready, Yana turns to ChatGPT and its image generation model (what she calls Images 2.0, which is DALL-E 3) to create a core product photo. She mentioned that she’s experimented with many models but finds that ChatGPT’s is the best at following visual directions, especially when she starts with a hand-drawn sketch. It adheres to the design's lines almost perfectly—sometimes too perfectly, making it look like paper, which is why her detailed fabric prompts are so important.

For the demo, she started with a text prompt based on her spec:

"So here we have a sharply tailored, waist-defined jacket with exaggerated shoulders and blah, blah, blah, right?"

Step 3: Iterating for Different Contexts

Once she has a core product image she likes, she doesn’t stop there. She uses it as a foundation to generate a whole suite of marketing collateral by asking for different styles and settings. She simply iterates on the initial image with follow-up prompts, often using voice.

To get a runway shot, she prompted:

"Make it into a Vogue style photo of a model walking down the runway in this outfit."

She does the same to create influencer-style photos and other visuals, which gives her a complete set of assets for her website and social media without needing an expensive photoshoot. It’s a fast, iterative loop that moves from a single idea to a full campaign's worth of imagery.

Workflow 2: Prototyping the “Unmanufacturable” Garment

Some of Yana's designs are so complex they would be nearly impossible to create with traditional methods. This is where AI moves from being an image generator to a production partner. She showed me an incredible dress inspired by the artist Ruth Asawa’s looped-wire sculptures, which is a perfect example of a garment that was previously unmanufacturable.

Step 1: Generating the Concept from a Photo

Instead of a sketch, this design started with a photo from her phone. She was repeatedly encountering Ruth Asawa’s art and felt inspired. So she prompted the AI directly:

"make a photo of a model walking down the runway in a dress inspired by Ruth, Ruth Asawa's, uh, loop wired sculptures."

She cleverly added that the model should have a “beige undergarment” to avoid the model's safety filters, which often misinterpret artistic images. It’s a practical tip for anyone working with image generation models.

Step 2: From Image to 3D Model with an AI Agent

To actually produce this dress, the spherical components need to be 3D printed. This requires creating precise CAD files. Instead of learning complex 3D modeling software herself, Yana uses OpenAI Codex with its “computer use” feature to act as an agent that operates the software for her.

She instructs Codex to use purpose-built programs like CLO 3D, a professional fashion design software, to create the necessary .stl files for 3D printing. This highlights a fascinating trend: SaaS isn’t dead, but AI agents might become the primary users. Codex on its own isn't great at generating a final CAD file, but when it’s driving software designed for that task, the results are excellent.

Step 3: Running AI and Humans in Parallel

Yana admitted one of the biggest unsolved challenges is getting AI to create accurate, production-ready sewing patterns. To tackle this, she’s running a fascinating experiment: pitting AI against humans. She has human pattern makers working on the designs while simultaneously tasking Codex with the same job. It’s a race to see which process gets her to a perfect, manufacturable pattern faster. This pragmatic, head-to-head comparison is a great way to push the boundaries of what's possible.

Workflow 3: Running the Business with an AI Technical Co-Founder

This is my favorite part of Yana’s story. For her, AI isn’t just for creative work; it's her entire technical and operations team. As a solo founder, this is the ultimate force multiplier, allowing her to build a business that would have traditionally required hiring a team of engineers and business development staff.

Step 1: Sourcing and Contacting Manufacturers

Once a design is ready for manufacturing, Yana needs to find the right production partner. She uses the “deep research” feature in ChatGPT to find vendors that meet her specific criteria, like US-based custom apparel manufacturers.

After identifying potential partners, she uses a combination of voice commands and the “browser use” feature to have the AI draft outreach emails directly in her Superhuman client. She maintains final control, however, by making sure she’s the one to review the drafts and click the “send” button. It’s a perfect human-in-the-loop system that offloads the tedious work while keeping strategic oversight.

Step 2: Building and Maintaining the E-commerce Site

Yana had no human engineers build her website, Yana Bana. She did it all with Codex. She simply instructed it to build the site, which is hosted on Vercel and managed through GitHub.

When she needed to add pre-order functionality, she just asked Codex to integrate Stripe. After a brief back-and-forth where the model initially claimed it couldn't, Yana insisted, and it completed the task. This is a common experience—sometimes you have to be a little persistent! She also had it build a database to track votes on which garments users want to see produced.

Step 3: Replacing Dashboards with Voice

Initially, Yana had Codex build a dashboard to track the votes from her database. But she soon realized she could create a more efficient, AI-native workflow. Now, she doesn't even bother with a visual dashboard. She just uses voice to ask her AI co-founder for the latest numbers directly.

"Oh, hey, um, how many votes do I have for this thing now?"

This simple shift perfectly illustrates the move away from traditional software interfaces toward more direct, conversational interactions with our systems.

AI as a True Enabler

Yana’s work is such a powerful demonstration of AI as a true creative and business partner. She has stacked these workflows together to create something that would have been incredibly difficult, if not impossible, for a solo founder just a few years ago. From ideation to production to sales, AI is the thread that connects every part of her business.

I’m so inspired by her vision and execution. I encourage everyone to check out Yana Bana to see the designs and vote on your favorites. You can also follow Yana's journey on X, where she is @yanabana.

***

Thanks to our Sponsors

• Merge: Merge is the infrastructure layer for production AI. It connects to thousands of tools, gives agents secure ways to act inside them, and optimizes model routing and spend without you building or owning any of it. Visit merge.dev/howIAI to start building for free.
• Jira by Atlassian: The teamwork graph in Jira delivers 44% more accurate agent results with 48% less token usage. The teamwork graph pulls context from across your entire stack — from Jira and Confluence to GitHub — and feeds it directly to your agents before they write a single line. Try them free at jira.dev.

Episode Links

• Yana Welinder on X: @yanabana
• Yana Bana (pre-order and vote on garments)
• Tools mentioned: OpenAI Codex, ChatGPT (Images, voice, deep research, browser/computer use), CLO 3D, Stripe, Vercel, GitHub, Superhuman