Training images for Explosive Situations, mostly made with Midjourney, but a couple were created with an earlier version of the LoRA from when it was trained on just 4 images and severely overfit.