Ganpan

At the core of the project is a digital database made by photographing and labeling individual letters on city signs. On the website built on top of it, users enter their own sentences, which are automatically assembled and displayed as images of letters extracted from the signs.

Ganpan not only stores the sentence images users create in digital form, but also turns them into analog publications in the form of zines. This brings the artistic process from the digital back into physical form, in people's hands, and explores how contemporary digital media can work complementarily with analog.

Through the visual reconstruction of text, users are invited to take a fresh look at the spatial landscape of Korea, while the project presents the possibilities of digital art (website) and print media.

Ganpan operates in the opposite direction of AI image generation. While AI learns from countless sources to create images with untraceable origins, Ganpan collects individual letters from Korean city signboards and clearly preserves their sources. User-inputted text is transformed into actual signboard letters, and clicking each letter reveals its original signboard context.

Figure 1. Diagram comparing an AI image generation model and Ganpan

Ganpan stores original signboard images and their extracted individual characters (e.g. '은', '마', '종', '합', '상', '가') in a relational structure within a Supabase database. When a user enters text, the system queries the database to retrieve matching letter images and displays them on screen. Each letter image is linked to its source signboard image through a parent key, so clicking any character shows the original signboard context.

Figure 2. Ganpan website's pipeline

More than 300 user-generated signboard images were collected in the dataset. I used three.js to create a plaza of these images, where you can see them combine to form a tall building.