Estudio Foto AI

An AI portrait studio promises magic, but under the hood it's a heavy process: upload images, fine-tune a model per person, generate hundreds of variants and deliver them. Doing that by hand for tens of thousands of users was impossible; the flow had to be 100% automatic, from the card to the result.

Training and generation take minutes, not milliseconds. The platform couldn't freeze while it waited: it needed to process in the background, notify the user when the work was ready and absorb spikes without dropping jobs.

Above all, it had to be profitable. Every unit of compute counts when you generate half a million portraits, so the pipeline had to be efficient and predictable.

We built the product in Next.js on Vercel and moved all the heavy lifting to Replicate: a dedicated model is fine-tuned per user and then the portrait batches are generated. The front end never blocks —it fires the job and moves on.

Generation is asynchronous: we queue each job and track its progress over webhooks, so the user receives their galleries the moment they're ready. Originals and results live in AWS S3; Firebase stores state, accounts and metadata.

Stripe charges before compute even starts, so every paid job triggers its own pipeline with no intervention. An integrated blog and marketing pieces feed acquisition for the 80,000+ users across the AI photo line.