Recovering Lost Digital Art
I had a thought in my sleep this morning. Becky Truman had asked me how she could get access to one of the images she painted on an experimental full color paint program running in Kajiya's Computer Science lab at Caltech in the 1980's. She had named the image "User Friendly", and I remember it.
The Technology
The Caltech Paint Program ran on a DEC PDP-11 computer connected to a Grinnell GMR-27 frame buffer, running a full-color paint program. Full color was very rare back then. Think about the displays on Mac and PC (which hadn't come out yet). They offered maybe 16 colors, if you got any color at all!
Yet we were painting in full-color (24 bits per pixel) at 512×512 resolution.
Expensive full-color paint programs like Lumena and Quantel became available in the mid-1980s, but they required special hardware. They were all replaced by Fractal Designer in 1991, which ran on commodity PC hardware, and eventually Photoshop.
The Recovery Journey
I dug through my storage and found an Exabyte backup tape. Nobody has used Exabyte 8mm tapes this century. I found Exabyte drives for sale on eBay, and bought one. I didn't think about it at the time, but the tape drive uses the SCSI interface, which was very popular—except that it hasn't been popular this century.
I considered going down that rabbit hole. Then my friend Al Alcorn introduced me to his friend Al Kossow, at the Computer History Museum. Al has Exabyte drives up and running, of course, but he couldn't read the tape. A few weeks later:
He sent me the entire contents in a file called "gumby.tar.gz". The computer's name was "gumby", thanks to John Snyder, now at Microsoft Research. The file expanded no problem. (Sadly, the archive is truncated, so we don't have all of it. But we did get the entire directory containing the paint program images.)
The .pix File Format
Now I had a bunch of images in a file format called .pix.
Keep in mind these files were created a long time ago—a decade before PNG, before JPEG, before TIFF, before GIF. There were no standard full-color file formats, only proprietary ones. At Caltech we used .pix format.
Back then, Lauren Carpenter and Rob Cook of Pixar spent a lot of time with us. The least productive collaboration we did, unquestionably, is that we debated image file formats. After that meeting, I wrote a program called "image" that used heuristics to read and write images in any file format including .pix (and do lots of processing in the middle, controlled by a command line interpreter—thank you Jim Blinn).
Unfortunately, "image" isn't in the backup, and even if it was, it wouldn't run and probably wouldn't compile. At least the code would act as documentation for the file format.
I spent an hour or two trying to decode the .pix files, but then I got distracted, and haven't gotten back to it.
Enter Claude
Until this morning. While I was asleep, it struck me: Claude!
I told Claude what I remember about .pix: that it is full-color, lossless, and uses run-length encoding. I dropped three images. Claude spun for about five minutes then displayed the images! It was that easy!
(Of course, something was off, and I told it to try harder, and then it got it right.)
Early Ray Tracing Research
The Grinnell frame buffer was also used to see the results of our early ray tracing research. Among the images we find Kajiya's first images of procedurally generated models:
The Magic Egg (1984)
Here are images of the space ship sequence that we contributed to the (unwatchable) 1984 Omnimax collaboration, "The Magic Egg":
Watch the film: The Magic Egg on YouTube
Resources
Claude conversation: Decoding the .pix files
Grok's interpretation: YouTube Short
Grok interpreted User Friendly this way. When I attempted to refine it, the results were blocked.