I've finished a proof-of-concept of the idea. Reused some code of boomrang.app to start with a Golang backend and a Typescript Next.js backend running on Vercel.com.
The difficult part of this project is integrating with a number of parties. A shop provider, a printing company, and a neural style transfer service.
Long term I want to bring the machine-learning model in-house to provide optimized models for wall-art.