Trained on 50+ HQ images of explosions. Add "Explosion Artstyle" to your prompt to make something blow up. Per usual you may need to fiddle with the weights of things to get the desired results, but for the most part it has a pretty strong influence. Works well w/ IMG2IMG also using medium value denoise strength.

Trained on V1.5 using Dreambooth w/ Joe Penna's repo (locally/Jupyter lab). Extracted from CKPT to LoRA using the A1111 Supermerger extension. Should work decently with any models based off of SD V1.5. (About half of the samples used Realistic Vision for example).

Just for kicks.