AnalysisVisual AISeptember 26, 2026

Krea 2 LoRA trained on 356 real portraits enables direct face-feature prompting

Read original source →reddit.com

A Reddit user trained a LoRA on 356 snapshot portraits of real people, captioning each with a multi-parameter face vocabulary so features can be prompted directly. The goal was to escape Krea 2's polished, inoffensively attractive default face for fashion shoots and editorials.

1 source

More stories today

Open the live feed