<h1 id="heading-2132">Son Mo-Ah | 손모아</h1>This model was aimed to produce images of Son Mo-Ah from "I stole the Number One Ranker's Soul", a manwha that i recently started reading and that i found interesting.There are 3 version of this lora:<ul><li>Soft, the one that received less training, but that overall can produce nice enough images.</li><li>Hard, this is a model that received some more training compared to the soft one. I think it's more flexible and also prefer it due to more consistent results.</li><li>Completed, this is the model i like the most, has a wide expression support, and i think gives a feeling of the serie style.</li></ul>The models work best with strenght between: 0.6-0.9It gives decent outputs with very few tokens (i used about 12 on positive prompt for the generation of the example images), but if you use few tokens you might want to use some negative embeddings like "easynegativev2".The activation keyword is:<pre><code>SonMoAh</code></pre>There are also some other keywords that might be affected by the lora:<pre><code>1girl, solo, long hair, brown hair, brown eyes, breasts, upper body, all_four</code></pre>Overall the lora is pretty flexible with expressions and poses, but you might want to inpaint some areas in the post processing to increanse the quality of the details (i suggest to set in-paint only selection, and a variation strenght between 0.6 and 0.65).N.B: i in-painted (using the previously mentioned method) only the face of the example images to increase the details, but the results where pretty ok even without it.

Files with SHA256: 9ce8f7947c0a35913609bdbdf6df46cedd1db4528b314fdea2c31adf41882f54