I've always found amusement in the goofy style of adult photography featuring exaggerated perspectives, where a woman's head with a startled expression is placed next to an oversized penis. I wanted to replicate such images using Control-Net or tags like "penis awe," but achieving precise results proved challenging. Therefore, I decided to train a mini-LoRA. While I'm not adept at creating complex LoRA models, it's essential to note that this may not function flawlessly every time. Nonetheless, I'm content with the outcomes of this conceptual model. I conducted training on equal amount of both male and female subjects, despite the scarcity of footage related in gay pornography. Users employing models designed for male-oriented content with pre-defined penises might encounter difficulties with the frenulum aspect in the glans of dick, but increasing the LoRA strength may resolve it.

This entire model underwent training using the default 1.5 Stable Diffusion model, rendering it highly adaptable to other models. Utilizing tags such as "detailed face, close-up" can enhance desired results. Although the primary focus of the model is capturing reactions to an enlarged penis, incorporating various expressions like smile, ahegao, sleeping, can be also beneficial. If the final output has glitches, adjusting the LoRA strength to around 0.5-0.9 may be able to resolve the issue. Additionally, you can even play around with this LoRA in img2img, if you wish to reposition the penis or something.