1. I treated the recording process as manual generative, IE typed ‘search’ into metacrawler’s image search with size specified as large and restrict to JPG files.
Metacrawler doesn’t serve them up as neatly as google so had to click through to the sites where i saved the the first recognizable image – this is perfect for automating via scripting, and i think could easily done through e.g. a flash interface
2. the images were all saved into a folder
3. ran a photoshop action magnifying the images, sans antialiasing and pasting them into a layered tiff file – 20 of them – into a 7087 px square tiff – this was SLOW
4. from here it gets a bit more ‘performative’ and I’m note sure how to translate it into our automaton (but have some ideas see A. below) – in that i was trying to find an emergent image in the layered synthesis so:
1. disabled all but the bottom two layers, setting the higher one to 50% – then manually rotating and scaling untill it I imagined I saw recognized something new in the image, then same with next layer at 33%, then 25%, then20%, these layers then merged, and the process repeated with the next batch of 5 layers, until it leaves just one layer.
5. finally i perform a kind of signiture filter effect that i do to most all these

A. Finding the emergent: say we make a copy of each source image, reduce it to binary Black/white, just ones and zeros. Then use a ‘little math’ to identify the biggest light area, and automate aligning/positioning this across the layers via scaling and rotation of each entire map, (which will invariably create some cropping artifacts that typically add to the cubist-look factor and fine )