Test revision of lora trained on set of underwater images including creatures, plants and surfaces. All images in dataset is of cc0 license. It's very first draft with 100 epochs and pretty bad captions.
Description
this is very raw/draft version, for now it gives some recognizable result with lora strength set from 3.5 to 4.5, cfg of 3 to 5, and 4-12 steps. keyword=jellyfsh