On digestion, going slow and whether the custom AI model still matters

The dominating discourse about AI — not only in art but also in the corporate world — still seems to be about its "generative" features: creating more images, using more tokens, more connectors, more compute, more scale.

I've always found myself drawn to the opposite direction: going inwards, and playing with the "digestive" capabilities of AI, going slow.

As an artist, I train a model on my own material — archives, childhood images, personal texts — and use it to digest an existence until something surfaces that I couldn't see before, but was indeed always there. This is what I call '“the light that is not seen”.

When I wake up, I describe my dream to my tailored text-to-image model; the machine offers image hypotheses in a breath, which can later be projected into virtually eternal mediums, such as pietra dura.

So I'm curious about your experience — as artists, curators, or simply enthusiasts:

Is AI more interesting to you as an engine of production, or as an organ of digestion — a way of seeing rather than a way of making more?

And does training your own AI model still matter, or does nobody really care anymore? Plenty of strong work now runs on off-the-shelf tools. I still train mine because it’s important for me to maintain a consistent visual ecosystem -and I enjoy doing this part- but maybe that's a conviction only I still hold.