<h1 id="ahegao-2kjya0kfz">Ahegao</h1>...but with that PFP style.This LoRA was trained on a dataset featuring anime girls with that ahegao facial expression, styled after those hentai profile pictures (PFPs).<h2 id="basic-instructions-iyztop8kl">Basic Instructions</h2>highly recommended with <a target="_blank" rel="ugc" href="https://civarchive.com/models/9409?modelVersionId=384264">Anything XL</a>Simply add <code>ahegao</code> to your positive prompt to generate a close-up of the face, often with details like a tongue sticking out or eyes rolling back. You can specify these elements if desired. Since the training captions used Danbooru prompts, using tags like <code>1girl</code> is highly recommended.make sure to use a 1:1 aspect ratio, or you might get distorted, stretched faces.I recommend using a strength of 0.8, but if you encounter weird artifacts like duplicate tongues or teeth, try lowering the strength by a tenth. It should help, but I can’t guarantee it—just cross your fingers and hope for the best!This LoRA was only trained on ahegao faces, so it’s not intended for anything beyond that.<a target="_blank" rel="ugc" href="https://civarchive.com/models/9409?modelVersionId=384264">Anything XL</a> is the best choice for this LoRA since it was specifically trained with that model and naturally has that 2D flat color aesthetic.You can try using other models like <a target="_blank" rel="ugc" href="https://civarchive.com/models/257749?modelVersionId=290640">Pony</a>, but the results will be poor since the model's prompting doesn’t align well with the LoRA. During testing, I achieved somewhat decent results by encouraging a 2D flat style, but they were still far from ideal.

Files with SHA256: eb165f78406300908675636d01e27666ce1ff0359028d4873a5b8c8c5f935c8c