User compares six base models for LoRA training

A Reddit user trained two faces on six base models (Ideogram, Flux.1 Dev, Flux.2, Klein, Krea, Z-Image) and found Ideogram 4 held likeness best. The experiment used RTX 4070 Ti SUPER cards and automated training with Claude.