Trying to make out paint for text (essentially a Large Symbol Model). Tokenize renders of PDFs with patches of some sort and see if I can make a multi language monochrome visual vocabulary where I can embed language and document metadata. I want to be able to convert my tax statements into musical compositions and render pride and prejudice as a 80s computer technical manual. Could also allow for direct language translation without the complex language tokenization we have today. Literally visual pattern understanding.