At a glance
WHAT IT’S REALLY ABOUT
How an AI-native fashion founder designs, sources, and ships solo
- Yana Welinder demonstrates how she runs an AI-native fashion brand (Yana Bana) by using ChatGPT/Codex across design, production, sourcing, and technical implementation.
- She relies on a detailed “fashion prompt stack” that acts like a specification for garments—covering silhouette, fabric behavior, construction details, movement, and other attributes—to generate consistent, controllable outputs.
- Yana prefers ChatGPT Images (Image Gen 2.0) because it follows sketches and visual directions more faithfully than other models, even if it sometimes overfits to sketch-like rigidity and needs fabric-flow guidance.
- For complex, previously impractical designs, she combines Codex with “computer use” and specialized software (e.g., 3D tools, CLO) to create CAD/STL assets and move toward real-world manufacturing.
- On the business side, AI handles vendor research, outreach drafts, browser-driven email setup, and building the e-commerce site with Stripe and voting databases—letting a solo founder operate with “technical co-founder” leverage.
IDEAS WORTH REMEMBERING
5 ideasUse AI end-to-end, not just for ideation.
Instead of treating AI as a single “design generator,” Yana applies it across the entire pipeline: concepting, visualization, production planning, vendor discovery, outreach, and web/ops. This turns AI into a practical force-multiplier for a solo founder rather than a novelty tool.
A prompt should function like a spec: define “good” in detail.
Yana built a reusable “fashion prompt stack” that encodes silhouette, proportions, fabric behavior, construction details, movement, and even sound. Claire frames this as the broader lesson: a prompt that behaves like a product spec/PRD improves results for both AI and humans.
Model choice matters: prioritize instruction-following over generic realism.
Yana prefers ChatGPT Image Gen (noted as Images 2.0) because it follows her visual inputs—especially sketches—more faithfully than other models that may look realistic but drift from the intended design. She also notes a common failure mode: over-adhering to sketch lines can make garments look like paper, requiring explicit fabric-flow instructions.
Pair agents with specialist software to ship real artifacts (e.g., CAD/STL).
For a Ruth Asawa–inspired dress, Yana describes a path from AI visuals to 3D-printed components, where CAD creation is the bottleneck. Codex alone struggles to output final CAD reliably, but combining Codex with “computer use” inside dedicated 3D/CAD tools produces workable STL files.
Offload vendor research and outreach while keeping a human “send” gate.
Yana uses AI to find manufacturers, run deep research, draft outreach emails, and—via browser automation—prepare messages in Superhuman while keeping herself as the final human checkpoint on sending. This removes the “boring middle” that often stalls creators from turning ideas into products.
WORDS WORTH SAVING
5 quotesNormally, I would hire a technical team, like engineers. I didn't, right? I sort of just came to Codex and asked it to build the website. The idea is to use AI really anywhere in the process where that makes sense. Literally my technical co-founder.
— Yana Welinder
It describes a lot of sort of what goes into a garment, so it has, like, the, the silhouette, uh, the proportion or the volume of, of the dress, like, kind of like the fabric, how, how it's, how it flows, how it behaves, um, the, the construction details, how it moves, um, even how it sounds.
— Yana Welinder
It's product people, yeah. It's a spec. You gotta have a spec.
— Yana Welinder
This garment is previously unmanufacturable.
— Yana Welinder
So I'd, I'd say it's kind of two things. One is that, like, getting, getting it to really do, getting image models to do really unique things per your vision while still making it realistic. That i- that has been really, really hard.
— Yana Welinder
High quality AI-generated summary created from speaker-labeled transcript.
