Sjors van Holst

After I finished my studies and continued work on Swiftbite, I was looking through the user feedback and noticed that a lot of users were very impressed with the icon set the app contained. I was glad to hear this, of course, as I had spent quite a lot of time setting this up.

When I first created the icon flow for Swiftbite, the image generation models were much worse. So I spent a lot of time setting up a fast Docker container and endpoint relay system, along with an image processing pipeline and a validation step to ensure the icons looked good. It took a lot of work to build. During my assignment, OpenAI released gpt-image-1, which made the entire pipeline redundant, as it could generate high-quality, natively transparent images that I could call directly from a Next.js endpoint. That meant I could throw out all my earlier work, which was a very bittersweet experience, even though the icons do look way better now.

I figured other food-tracking apps or services might want to add something similar. The setup I created wouldn't be very hard to recreate for another service, but generating all the icons I already had does cost some money, so I thought I might as well make it available to other people. This way, they only had to add an img tag with the right URLs, and they'd be off to the races.

So, I turned the icon logic of Swiftbite into a standalone microservice. I created a landing page with some examples, a demo, and very simple API documentation. For now, I consider the service to be in beta, so to obtain an API key, users can contact me and I will provide it!