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):<pre><code>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, polynesian</code></pre>Note: no need to use generalized and specific triggers together (asian and asian_chinese for example)<pre><code>The lora can be used without triggers as a style lora (example image with tifa, aerith and yuffie in a group hug)</code></pre>Currently the lora supports the following races &amp; ethnicities:E.Asian (Japanese / Korean / Chinese)European (General European / Irish / french / german / italian)African (General african / east / west)LatinaThaiIndianIranianArabIndigenous American (as of v3.3. Native American in older versions)PolynesianThe Above list is for version 3.3Please refer to the list of triggers for the supported tags for older versions<pre><code>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.</code></pre>

Race & Ethnicity helper

New:<ul><li>african_east, african_west (Couldn't find nearly enough images to separate into countries so it's just east and west for now).</li><li>Added more images of males to the dataset so it should be easier to generate males</li></ul>Changed:<ul><li>'native_american' changed to 'indigenous_american' to reduce mixing with the Danbooru tag.</li><li>Latinas are now fox girls (no idea why, for now put fox ears/ animal ears in the negative).</li><li>Iranians more likely to have pink hair (same, can be prompted out)</li></ul>It's trained on largely the same dataset, but the tagging method has changed. Mostly an improvement in anatomy, and skin quality. A dedicated pony version will come later.

Files with SHA256: 8d2e83910481b9de93c3e87246da77fa3eabf2ec8a4418c1c36de3c2c2eefd02