Fashion and e-commerce teams needed campaign imagery at scale — layflats, packshots, and phone shots turned into on-model, lifestyle, and marketing assets — without a photoshoot for every variant. Flat per-run pricing, brittle uploads, and a single-cloud storage model would have made the product untrustworthy or unbillable.
The Solution
I shipped production surfaces across the generate loop: React tool UIs, internal job/credit APIs, and the systems that make AI output operable. Work included formula-based pricing aligned on frontend and backend, HEIC and oversized-upload handling, GCS/S3 storage abstraction, and credit-aware UX so teams see cost before they burn a generation.
Key Features
AI workbench UX: workflow selection, upload or reuse library assets, live credit cost, submit → poll → review → regenerate, with insufficient-credit and zero-credit states.
Dynamic pricing: static, per-duration, tiered, and per-megapixel formulas plus stacked multipliers (resolution, duration, aspect ratio, audio, formatting), mirrored in the UI and billed the same way on the server.
Upload reliability: HEIC/HEIF ingest, auto downscale for large megapixel/edge limits, compress under the processed-size ceiling, EXIF orientation fix for iPhone photos.
Storage interface: GCS and AWS S3 behind one contract — writes to the active provider, reads by object scheme — so libraries survive provider switches.
Monetization integrity: credit grant/consume/refund/expiry paths and failed-job refunds that stay correct under queue retries (implementation details under NDA).