Fashion DSMS - the Digital Sample Management System

Inspiration

The fashion industry is one of the most creative industries on earth — and somehow one of the most operationally broken.

We kept coming back to one absurd fact: a brand designs a garment digitally, then pays a factory thousands of dollars to make a physical version of it just so people can look at it — and then throws it away. This process repeats six to ten times per style, across hundreds of styles, every single season. It is slow, expensive, wasteful, and completely normalized.

When we looked at Perfect Corp's visual AI capabilities — the body detection, the skin and fabric rendering, the try-on infrastructure — we saw something the fashion industry desperately needed but hadn't connected yet: the ability to generate believable, body-accurate product images from nothing more than a spec file. The rendering problem was already largely solved. What was missing was the workflow around it — the collaboration, the versioning, the approvals, the supplier communication — that would let teams actually replace their physical sample process rather than just supplement it.

That gap became Fashion DSMS. We didn't want to build another pretty try-on feature. We wanted to build the operating system that makes the physical sample optional.

What It Does

Fashion DSMS is a collaborative digital sample management platform that replaces physical garment samples with AI-generated product images and a structured approval workflow.

A designer uploads a tech pack — a standard spec file they already produce — and the platform generates a photorealistic image of that garment on a realistic, diverse human body, powered by Perfect Corp's visual AI engine. From that point forward, every stakeholder in the product creation process — designer, technical team, factory, buyer, and brand director — works on the same living digital sample inside one platform.

The core capabilities include a digital sample creation pipeline that converts uploaded specs into body-accurate garment images across multiple body types and colorways. A collaborative review layer allows stakeholders to annotate directly on the image, request revisions, and track every change with full version history. A supplier and factory portal gives manufacturers a structured channel to receive briefs, flag construction concerns, and confirm approvals without a single email. A sample lifecycle tracker manages the rare cases where physical samples are still needed, keeping digital and physical workflows unified. And a decision dashboard gives leadership a real-time view of where every style sits in the approval pipeline, projected costs, and time-to-market status.

The result is a process that compresses weeks of back-and-forth into hours of collaborative digital iteration — before a single thread is cut.

How We Built It

We built Fashion DSMS on top of Perfect Corp's visual AI and try-on API as the core rendering engine, using it to place garment imagery accurately on diverse body types with realistic fabric representation.

The frontend is built in HTML with a component architecture designed around the core workflow states — creation, review, approval, and production handoff. We used a canvas-based annotation layer so stakeholders can pin comments directly to specific parts of the garment image, similar to how tools like Figma handle design feedback but purpose-built for product review.

The backend is a PHP REST API managing version control for digital samples, user roles and permissions across brand and supplier accounts, and workflow state transitions. We used MySQL for structured sample and approval data and AWS S3 for image asset storage and versioning.

The tech pack ingestion pipeline parses uploaded spec files and structured inputs, extracts garment attributes — silhouette, fabric type, colorway, key measurements — and passes them as structured prompts to the Perfect Corp API to generate the initial product image. Subsequent revision rounds use the same pipeline with updated parameters so every render is traceable back to a specific set of design inputs.

We built the supplier portal as a lightweight separate interface so factory partners can participate in the workflow without needing full platform access, keeping onboarding friction low on the manufacturing side.

Challenges We Ran Into

The hardest technical challenge was translating a flat, text-heavy tech pack into a structured set of visual parameters that the rendering engine could act on reliably. Tech packs across the industry are inconsistently formatted — some are detailed CAD exports, some are rough PDFs, some are hand-annotated spreadsheets. Building an ingestion layer flexible enough to handle that variability without breaking the downstream render pipeline took significant iteration.

The annotation layer presented its own complexity. Pinning a comment to "the left shoulder seam" on a rendered image sounds simple, but when the same garment is rendered in four colorways across three body types, that comment needs to persist and display correctly across all twelve images simultaneously. Getting the coordinate mapping right across different image dimensions and crops was tedious but critical to the workflow feeling seamless.

On the product side, the challenge was scope. The full vision for this platform is enormous — cost engineering, sustainability scoring, ERP integration, factory logistics tracking. Deciding what the core hackathon prototype needed to demonstrate to be compelling — without overbuilding — required real discipline. We kept returning to one question: what is the minimum workflow that proves the physical sample is replaceable? That became our north star for scoping.

Accomplishments That We're Proud Of

We are most proud of the end-to-end workflow functioning as a single coherent experience rather than a collection of disconnected features. From tech pack upload to rendered image to annotated feedback to revised render to approval sign-off — the loop closes entirely within the platform. That full cycle, working in a live demo, is the proof of concept that matters.

We're proud of the annotation system. The ability to pin a comment directly to a part of a garment image — and have that comment persist through revision rounds — is something that sounds small but fundamentally changes how product review conversations happen. It removes ambiguity from feedback in a way that email and Slack simply cannot.

We're also proud of how naturally Perfect Corp's API fit into the workflow layer we built around it. The integration feels native, not bolted on. Their visual AI handles the hard rendering problem; we built the collaboration, version control, and business logic on top. The division of responsibility is clean and the combined result is something neither could be alone.

What We Learned

We learned that the fashion industry's biggest technology problem is not capability — it is workflow adoption. The tools to generate good garment imagery have existed in various forms for years. What has prevented adoption is that those tools don't fit inside the process teams actually use. A render that lives outside the approval chain, outside the supplier communication, outside the version history — is just a pretty picture. Embedding the render inside the workflow is what makes it a sample replacement rather than a sample supplement.

We also learned how deeply the physical sample is entangled with trust. Brands don't sample just for visual reference — they sample because they don't fully trust that what they designed is what the factory will produce. Any system that wants to replace the physical sample has to also replace that trust mechanism with something more reliable — which is why the version history, the structured factory brief, and the approval audit trail are not nice-to-haves. They are the product.

What's Next for Fashion DSMS

The immediate next step is deepening the tech pack ingestion pipeline to handle the full range of formats the industry actually uses — including direct integrations with major PLM platforms like Centric and Browzwear so brands can pipe design data directly into Fashion DSMS without any manual upload step.

The second priority is expanding the body model library to include a full size range with diverse body compositions, ages, and adaptive body types — including wheelchair users and limb-different bodies — since this is a massively underserved segment of the market that existing tools ignore entirely.

On the intelligence side, we want to layer in AI-assisted fit flagging — where the system automatically identifies potential fit issues (pulling, coverage gaps, proportion problems) before human review — and a wear-frequency predictor that uses historical return data to flag design choices correlated with poor post-purchase satisfaction.

The longer-term vision is becoming the system of record for physical product creation — not just apparel, but footwear, accessories, and eventually hard goods manufacturing — with a full API layer that allows any brand to embed the digital sampling engine into their existing tech stack.

The physical sample had a 100-year run. Its time is up.