Make sure to use UI that supports v-prediction, otherwise you will get a noisy blob!
Main parameters:
Steps: 28
CFG: 3-5
Sampler-scheduler: euler/euler a-normal, DDPM-exponential (better noses)
CFGZeroStar and sometimes Self-Attention Guidance: default for difficult backgrounds
Try to BREAK prompts into parts by 75 tokens max, this goes for most models..
Positive prompt:
number of characters (1girl/1boy), character name, artist prompt, scene/environment/camera angle, action, expression, items, quality prompts (masterpiece, best quality)some heavily stylized characters might require prompt: nose, realistic.. you can put nose into Adetailer
Negative prompt:
lowres, worst qualityeven longer negative should work, but I prefer simplicity.. tags worth noting include sketch, abstract, url
Enjoy..
Description
Should be very similar to the NAIXL, merged at 0.66 weights while it was fine at 2, HEHE BOI. I'm kinda happy for this version, I should expand my dataset (only 131 images), but i wanted to nail the kohya ss settings first. Anyways, training on 16GB is quite limiting, if you know how to do dreambooth with it DM me. :3 Prompts like good hands affect style as well as harsh lighting, adjust their weights to your desires.