June 24

Continued experimentation with generative video made several limitations clear. The outputs consistently leaned toward spectacle. Shipwreck imagery became aestheticised very quickly dramatic skies, collapsing vessels, cinematic framing. The results felt "cute," overly literal, and resistant to ambiguity. The framing with prompts drawn from cinema is interesting though. It's also kinda in the wrong point of the uncanny valley, I think I liked it more when it was consistently producing the equivalent of extra fingers, but the outputs are weak.

More practically, the economics of video generation made sustained experimentation unrealistic. The level of production I need just to figure out whats going on, essential for research, is just too high in cost and rendering time. I've spent more time trying to force out something I'm intrested in rather than making things.

At the same time the most interesting parts of the process are happening elsewhere: in text fragments, datasets, and procedural experimentation. The project is increasingly in scripts, folders, and partially structured information rather than images.

I'm not really sure how to proceed. The original proposal centred on moving image, yet the momentum is moving away from it.

I'm focusing on generating plausible maritime incident descriptions using structured language derived from Lloyd's reporting styles. The outputs are modest but unexpectedly convincing.

Without images, the events feel not so much less fictional, but more weird.