Quattro (nanoha) revised edition

This is a revised version with some prompt changes and increased training data. When using LoRA, it might be best to use LoRA 'DMD2' in conjunction with it, and generate the sample with Sampling Steps around 10 and CFG scale between 1.1 and 1.5 (other advanced settings can be left at their defaults). Adding the artist name to the prompt might also be effective, but you'll have to experiment with that yourself.