Experiments in diffusion AI: in this series I’m forcing/hacking a small locally-run diffusion model to produce abstract, averaged and rough images for a minimalist prompt, int his case “river”.
For the ones interested in the details: this was done in ConfyUI, with a small Stable Diffusion model, and reducing the steps, manipulating the noise values and scale, and cfg parameter. I decrease the values of noise and steps until all figurative traces disappear, then play with cfg values and minor changes to scaling noise until I get results that I like.