My first detection model! YOLO for mastectomy / breast removal scars.

Recommended strength: 0.5 or higher, trained resolution: 768x768

To use, add an ADetailer node or plugin to your UI and add it to its models folder. restarting may be required.

Note:

Since this YOLO model is best used with a LoRa, you may have discoloration when using a Lora and Adetailer together.

You can combat this by 1, check your VAE settings, or 2: Add your prompt without the lora annotation to the ADetailer prompt box. This means the lora will be skipped when inpainting it.

Notable problems:

1. Fold-over / dents: try adding "nipple indents" to your negative. This download is the 'last' model file where the 'best' model file (thanks YOLO naming scheme) had longer scars but much worse indentation.

2. Dark fur colors, say with a bear, don't seem to trigger at all? I did notice some pre-made datasets to test with have a dark and a lighter image. But this would double the work. 450 images, all hand tagged, times two. Bleh. A thing if this gets enough use, maybe. :P