
Build
A womenswear brand in India
Apparel and textiles
Generative-AI product imagery for a saree and womenswear label: garment-on-model composites, catalogue stills and a short fashion film, produced from flat garment photographs without a studio shoot.
Nothing was measured after delivery. No one set up the measurement, so the honest answer is that the outcome is unknown.
The problem
A small womenswear label photographs its stock the way most small labels do: the garment flat, on a hanger, under whatever light the room has. Those images sell the fabric and nothing else. A saree on a hanger is a rectangle of cloth; on a person it is a drape, a fall and a silhouette, and the second thing is what the customer is buying.
The usual answer is a studio shoot: a model, a photographer, a stylist, a location, and a cost per garment that only makes sense if the garment is going to sell in volume. For a label with a long tail of designs, each in small quantities, the economics never close. Most of the catalogue stays on the hanger.
What I did
I built the catalogue imagery generatively instead.
- Garment-on-model composition. Image-to-image generation that takes a photograph of the actual garment and a model reference, and returns the garment worn and draped. The constraint that matters is fidelity: the output has to be the garment the customer will receive, not a plausible garment of the same colour.
- Catalogue stills. A consistent set per design, so the grid reads as one brand rather than as a collection of experiments.
- A short fashion film. Generated motion from the same stills, for the formats where a still does not hold attention.
The work ran across two generation tools because neither does both halves well: one is better at holding garment detail through a composite, the other at producing a usable motion sequence from a still.
How to read this
This is a production method, not a measured outcome. What it changes is the cost of putting a design in front of a customer wearing it, which for a long-tail catalogue is the binding constraint. Whether the images sell better than the hanger shots is the label’s measurement to make, over a season.
Where it needs care
Generated imagery of clothing sits close to a line. The garment shown has to be the garment shipped, and the model has to be understood as a representation rather than an endorsement. Both are solvable with disclosure, and both are worth settling before the first catalogue goes out rather than after.
Other build work
- A scientific publishing services firmTwelve ranked automation opportunities, one shared architecture, and working prototypes: a reference and retraction checker, a books project-management system, and a rules-first manuscript-to-XML pipeline.
- An engineering design firm for building servicesFour hours per floor go into cleaning an architect's drawing by hand before engineering can start. I built and measured a prototype that does the tractable parts of that, and reported the part that is not solved.
- InticoreA quotation-rules knowledge pack for a custom GPT that reads a casting drawing, tabulates every dimension and tolerance, checks feasibility and writes the queries back to the customer.
Have a problem worth measuring?
I take a small number of advisory and build engagements each year.