I found Sava Schultz on TikTok and thought she was pretty. So, here's a LoRA of her. Using <code>bad_prompt_version2</code> with this LoRA gives the result a more stylized look. It also tends to give her hair more highlights. Just an oddity I've noticed. As for regular usage, a strength of 1 is best for semblance. You might also want to use this LoRA at lower strengths to make the subject more curvy. <code>cleavage</code> will make the clothing more revealing (or less when in the negative prompt). 25 steps work well, but you may want to go closer to 50 if you are having issues with the clothing. The thing about it is, that more steps tend to stray further from her facial distinctions. Now, I know the example images do stray from similarity a bit. To an extent, that is the base model, epiCRealism, showing through. Prompting more for clothing the LoRA likes gives better results, but wanders outside of the guidelines. For that matter, the one of her in a jacket is pending review as of writing (yet it is the most modest). Some of these LoRAs just don't like to cover up.

Files with SHA256: 11a2a379c36327ddca4af3cc37849f2d0db5abd4deba2e406345b15e312020dc