So here is a new model dedicated to the Khajiit from the game Skyrim. As for image generation, everything is not clear here. The problem with getting good image quality is the initially shitty training. Unfortunately, it is difficult to find adequate images of Khajiit, even on R34, which causes learning problems, since a total of 27 photos were used. However, if you make a double upscale (first by 2, then by 1.5), then you can get an adequate picture. As for the weights in the generation, I did not notice the difference, but I use 0.7 myself.As an example, I give several generations. One with my model, the other without. The seed is the same.First one. No model, no upscale: <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f923eaa3-c1a7-41c7-82cd-3c43c33a12ab/width=525/f923eaa3-c1a7-41c7-82cd-3c43c33a12ab.jpeg" />Second one. My model, no upscale: <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/4032ba24-88cb-4637-8332-a3c72bc422e8/width=525/4032ba24-88cb-4637-8332-a3c72bc422e8.jpeg" />As you can see, there are problems with rendering the face, and they are specific. However, as I wrote earlier, take a double upsale photo. As an example, the result of the upscale:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2e1f18b0-430c-40d1-ae62-21bde9622690/width=525/2e1f18b0-430c-40d1-ae62-21bde9622690.jpeg" />Will this model be more developed? Yes, it will be, since the result of generation is not very good and requires additional training.

Files with SHA256: ac80ad0131d10c53ef06b04157110f4448159e854ccb21f8dbbcaa636a71d773