Dreamyvibes Artstyle Alternate is a followup to a the "Dreamyvibes Artstyle" LoRA I uploaded a few months back. This version was trained from scratch using many well captioned images from the original dataset, as well as hundreds of interesting images that were generated w/ that prior LoRA.

Trigger with "Dreamyvibes Artstyle"

It should be more versatile than the original LoRA due to the training, but I can't say it's necessarily a progression or upgrade to version 1 - it's just an "alternative take" on the concept. From a training standpoint it's definitely much more complex than the original which should allow for more varied/nuanced generations.

Curated/captioned dataset was trained in stages on base SDXL w/ Onetrainer (Nerogar avail on github.com). Afterwards, this LoRA was extracted from the trained checkpoint using Kohya's extraction script. Captioning was done using GPT4 and Kosmos2.