This Lora is trained on mostly real images and a few anime/illustration images (~99% to ~1% ratio) to keep it flexible enough to still generate in anime style.
Triggers (as of 3.3):
asian, asian_japanese, asian_chinese, asian_korean, european, european_irish, european_french, european_german, european_italian, african, african_east, african_west, latina, thai, arab, indian, iranian, native_american, polynesianNote: no need to use generalized and specific triggers together (asian and asian_chinese for example)
The lora can be used without triggers as a style lora (example image with tifa, aerith and yuffie in a group hug)Currently the lora supports the following races & ethnicities:
E.Asian (Japanese / Korean / Chinese)
European (General European / Irish / french / german / italian)
African (General african / east / west)
Latina
Thai
Indian
Iranian
Arab
Indigenous American (as of v3.3. Native American in older versions)
Polynesian
The Above list is for version 3.3
Please refer to the list of triggers for the supported tags for older versions
For version 3.3 (both Pony and Illustrious), I decided to try frequency-based tagging instead of the common "prune anything you want the model to generate".
So basically any tag that appears with less than 20% frequency gets pruned, as well as anything above 95% and anything that's insignificant enough to be considered as 'noise' such as 'mole', 'mole on breast', mole on face', etc.
It does make the model slightly less flexible, so possibly pruning under 10/15% could be better.Description
Improved everything across the board, changed the east asian race triggers:
Japanese > asian_japanese , Chinese > asian_chinese , Korean > asian_korean
Extra note: I originally trained this as version 1.85 so that's whats in the preview images, I renamed it to 2.0 before posting
Update: changed the min strength on Civitai to 0.1 from 0.7 for the on-site generator.
It's supposed to help make characters look more like their race/ethnicity in anime format.
It can obviously generate realistic images (since it's trained on 100% real images) but it's not what I trained it for so I don't test that.
Because it's trained on real people it makes images look more realistic, but using "score_9, source_anime" seems to be enough in most cases to prevent that (worst case scenario you can add "realistic, photorealistic, 3d" to your negatives)Edit: the above refers to Hassaku XL and AutismMix, from testing with the onsite generator other Pony checkpoints may require the full string "score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, source_anime" and full negativeAt the moment the lora can generate white, black, asian (separated to japanese, chinese and korean), latina and thaiTrigger guide:
African: tends to generate dreadlocks and more muscle
European: white european, tends to make blondes and red heads, biased towards blue eyes if color not prompted
Asian_Japanese: closest to regular 2d anime style
Asian_Chinese: more slender waist, smaller heads
Asian_Korean: slender (more than Japanese, less than Chinese), wider hips, smaller heads {bias towards long hair, pale skin with a pink-ish tint, slightly more toned}Latina: brown skin, more curves
Thai: generates a mix between chinese and european.
It's possible to use multiple triggers to make mixed-race charactersFAQ
Details
Files
Available On (16 platforms)
Same model published on other platforms. May have additional downloads or version variants.