Castform

Problem

A fashion company wanted to let its own clients generate model imagery for their collections, without a photoshoot each time. The hard part was not calling an image model, it was turning that into a product: clients pick from available models, generate images for their own collection, and stay within whatever their plan allows.

What I built

• A guided generation flow: the client selects a model and generates imagery for their collection, so the AI sits behind a clear product rather than a raw prompt.
• Limits that hold: caps on images and on models per client, enforced in the backend so usage stays inside the plan and costs stay predictable, which is exactly where AI products leak money when the limits live only in the interface.
• Generation through OpenAI, integrated so the app orchestrates the calls, the results and the client's library of generated images.
• Plans and access, so different tiers get different limits, tied to billing.

Results

The interesting problem in almost every AI product is not the model, it is the metering: who can generate how much, and how you stop a good feature from becoming an unbounded bill. Putting those limits in the data and the backend, not the buttons, is what makes an AI feature into an AI product.