April 2024, starting.

Setting up: I'm setting out to explore generating narratives, the focus right now is getting up to speed with what systems I need, and start experimenting. Thats going to drive the next set of choices. I'm intending to create visuals, really to see what a system thinking of shipwrecks is, to elaborate narratives around the wreck of the Torrey Canyon in 1967 in the context of deep time. To see where this leads.

Early testing is about understanding prompting video and image systems. Prompt writing is maddening, this dance back and forth between specificity, then too much information and the context window seems to collapse. What I'm getting is unstable and theres a kind of boringness to it, it all looks somehow imagined by Disney. I'm not looking for realism, but the kind of not-realism I'm getting it's kind of imagine-less, understandably perhaps. Outputs drift toward something cinematic but oddly sentimental.

The images are interest at first glance but increasingly felt illustrative. closer to visual metaphor than material investigation.

Alongside these experiments I theres the broader research into maritime records, Lloyd's casualty reporting formats, shipping databases, and historical documentation practices. I'm planning on using this to see how the machine will elaborate from a stable set, I'm interested in the kind of institutional poetry of the records.