banner ad archive

A staple of the early 2000's internet was the humble banner ad. I remember seeing brightly colored ads on Myspace beckoning me to punch out George Bush to win an Xbox 360 or asking me install the latest custom cursors or emoji packs. Well before embedded video ads were feasible, Shockwave Flash ads were all the rage. They relied on flashy animations and sometimes audio to draw your attention to them and encourage you to interact with them.

In an effort to relive the nostalgia of these ads, I attempted to (with limited success) pull snapshots of Myspace from archive.org and extract banner ads from them. That's when I stumbled upon the Flash Banner Ad Time Capsule collection on archive.org uploaded by "Sketch the Cow", containing almost 6GB of flash ads from 2003-2004. 

If you're in the tech world, you may know that Adobe discontinued flash in 2020, which made previewing and sorting through these files a bit tricky. Ruffle offers an easy way to play them, but what if I wanted to (for instance) quicky preview multiple files or find ads for a specific website/company?  Then I thought of a plan: convert these files into GIFs and use an LLM with image recognition and OCR capabilities to process them. In short, I plan to perform the following steps across all of the tens of thousands of flash ads in the above dataset:

• Extract their frames using Ruffle's exporter functionality
• Use ffmpeg to convert the frames into a .gif
  
• Rename each file to its SHA1 hash and remove duplicates
• Use OpenCode and the MiMo v2.5 model to extract text and create keywords relevant to the ad
• Put this information in a searchable database
• Host it somewhere (?)

More to come on this project soon.