Falling Sky
Week three Summer into AI punk themed data discovery around falling meteorites.
This is my third submission for data punk Summer into AI. You can read about this week’s challenge here
Falling Sky uses the NASA Meteorite Landings dataset, discovered through its Data.gov catalog listing. Data.gov served as the public open-data reference point, while the actual meteorite records came from NASA’s hosted CSV, with fields like meteorite name, class, recovered mass, fall/found status, year, and latitude/longitude. The project turns those records into a 3D globe where each geocoded meteorite becomes a clickable archive signal, with mass-scaled bars showing how large the recovered object was.
The app was built as a static web project using HTML, CSS, JavaScript, Three.js/WebGL, preprocessed data assets, GitHub for source control, and Azure Blob Storage for deployment. I created it remotely with Hermes through Discord: I gave design direction and feedback in the thread, while Hermes edited the files, tested locally, deployed updates, verified the live site, and pushed changes to GitHub. We, Hermes and I, originally had a different custom visual direction, but the system flagged it as being out of alignment with the Data.Punk theme, we reworked the interface and hero treatment to better match the competition brief while keeping my original 3D visualization and art concept intact.
The globe rendering was an idea that I had done years ago at a creative agency that I worked at with some really good designers that I miss dearly. I recalled using billboarding line renders Unity 3D to achieve the vertical column effect scaling out from the Earth to represent each data point. I don’t know why I remember that all these years later, but I simply asked Hermes to implement that and it did and it worked well!
An unexpected lesson that I’ve had fun with was the new image editing via annotation inside of the desktop ChatGPT app. It’s extremely useful and does a very good job of editing little things. You can just click on a spot of the image and describe what you want changed about it and it just sends that in to the model with an X percent in a Y percent of where you clicked and the annotation, the model will re-render your image and it does a pretty good job of leaving the rest of the image intact. You can perform several annotations within one shot.
This being my third submission for this theme, I’ve made three hero or banner images in this way. A very useful prompt to get around a lot of AI over detailing images and general messiness is to ask for a healthy amount of negative space.
Needs an option to download the worldview as iOS background photo!