How would we navigate mobile applications without icons? From the timeless save icon to the controversial (in UX circles, anyways) hamburger menu icon, they help us instinctively know how to navigate almost any task or process.
For the third project in our ChatGPT Hackathon, Ryan Sweny and Frank Calpito devised an idea to simplify icon creation using generative AI.
While many icons might be universal, their design will vary from project to project. Whether you’re prototyping a new project or updating an existing one, it takes a lot of time to create a set that matches the look and feel of your design and the theme of your application.
This challenge inspired Icon Generator — an OpenAI-powered application that creates an icon library based on your inputs. Like the idea for Writing Assistant, Ryan and Frank built Icon Generator as a mobile application.
Icon Generator has three inputs for the user. The shape input lets the user choose between round and square icons. The mode switches between colour and monochrome icons. Lastly, the topic input is where the user enters the prompt for what icons are needed.
Once the inputs are set, Icon Generator sends the query to OpenAI and returns image sets to the user. The user can then choose one of the sets and generate new icons using the set as a base.
“The idea here is to keep clicking on generate and save whatever ones you want,” says Frank. “Then, later on, you can do a batch download when you’re back at your desk.”
Building the app
Icon Generator uses the OpenAI API to power the icon generation. If you recognize the name OpenAI, it’s because they are the people behind ChatGPT and DALL-E, a generative AI image generator. We wrote about experiences DALL-E on our blog, including how the tool is trained and what data is used for that process.
One of the challenges with the images created by the current API version is that it cannot produce readable text. Unfortunately, the API doesn’t allow you to exclude text, and some results included unreadable text. Some of it was pretty funny to try and read, though.
What we learned
According to online discussions, new versions of image generation AIs like DALL-E and Midjourney should soon be able to generate text along with images. Joe Reda, BitBakery CTO, said the great thing about building with an API like the one from OpenAI is that the Icon Generator app will get better as the API improves. He added that the Midjourney AI often produces better results, but no API is available.
“There's no reason why OpenAI can't improve as well,” says Joe.
For the next steps, Ryan and Frank are looking into different ways to save the icon libraries the application generates. They’re also investigating adding more prompt options to help users refine the icons.
Want to learn more about how you can add generative AI to your solution? Get in touch with us today to learn more.
Photo by Harpal Singh on Unsplash