This YOLO adetailer model is a fine-tuned version of YOLOv8m and is trained specifically to identify full penises in generated images (and avoid only partial ones). This allows the model to correct any artifacts and generate perfect penises every time.This was built specifically for use with my <a rel="ugc" href="https://huggingface.co/throaway2854/Futana-v1">Futana</a> line of Stable Diffusion models.It was trained using my <a rel="ugc" href="https://huggingface.co/datasets/throaway2854/Penis-Adetailer">Penis-Adetailer dataset</a>.

YOLO Penis Detection

v1 excels at penis detection but still shows some false positives where it detects penises not fully out. For now, I’d recommend setting the detection threshold to ~0.8 to correct for this. Future iterations will fix this but I was happy enough with the model as is to release it.

Files with SHA256: 7592b6b5ae0c702331a9fcfb34e6a6a84eaaea6645b8f37116b3bd5745266e35