Gathering Data
To collect our data, we used the Postcard Collection collection in the Carleton College Archives, found on the Digital Collections page.
We gathered 9 different postcards between the years 1900-1950. We each researched 3 postcards, evenly dividing the year range into 1900-1910, 1910-1920, and 1920+. With these postcards, we re-colored the original image with AI and rephotographed the original image location. Additionally, we researched the history of each building.
Why 1900-1950?
In the Postcard Collection, the scope of this collection only includes postcards from the first half of the 20th century. Speaking with the Carleton College Archives, we were told that these parameters were placed because they enabled the project to become more manageable.
Postcard Dataset
Methods
With the data that we collected from the postcard collections, we stored everything on Google Sheets to have easy access and keep track of our work. To show the viewer what images we have chosen, we created a past/present section for the viewer to get a sense of the building through the lens of the postcard and how the building looks when rephotographed as of Fall 2025. From the dataset, we made three visualizations for viewers to interact with, a map, a timeline, a 3D models.
We created our map with ArcGIS Online, which enabled a viewer to see where the image was taken, how much the image captured compared to the actual building, and what has happened to the building.
Our timeline was made with Timeline JS powered by Knight Lab to give the viewer knowledge when these images were taken according to the Digital Collections Year of Photo Taken category, the image of the postcard, and a description of the history of the building.
The 3D models of Williams and Gridley were generated by Meshy AI. By prompting it with one photograph, it was able to generate a cohesive model. We screenshotted a few angles of the model and removed their backgrounds. Then, we imported those images into Procreate, where the photo was place on top of a photo we took of the location. From there, the photos were manipulated and merged together to create a scene of what demolished buildings would look like if they were around today.
Map and Timeline Datasets
Use of AI
Colorization
While colorizing our images, we used DeepAI. However, we played with ChatGPT to colorize some of our work and compared the two. We noticed that ChatGPT took more creative liberty in its colorized version and decided that we would prefer to showcase our colorized version as naturally as possible.
3D Modeling
While efficient, using AI for 3D modeling has its limitations. It completely guessed on the building’s colors and back view. While these models may seem realistic, upon closer inspection they are not entirely accurate.
Presentation
With the map, timeline, and 3D models, this project applies methods of the digital humanities to understand Carleton’s history. Our website brings together postcard-based visualizations to offer a distinctive view of the campus’s past.