Can be run at both high and low strengths and produce decent results depending on what you're doing. SDXL already knows of this classic car, but the lora should help it stay consistent in terms of the shape of the car etc. I didn't train the lora on images of the finer details like the emblem, doors, interior etc. If there's a bunch of crap on the roof try writing "taxi" in the negative prompt.

Likes and posts to the gallery are much appreciated.

Why a new version so soon?
I went over my files and realized that something strange had happened - many of the .txt files containing the captions were missing. And as I looked at the files I realized that I hadn't been as thorough with my dataset as I thought, there were some variations with how the headlights looked etc. Whoops! This new version should perform much better.