This LoRA was made using the Kohya_ss flux finetunning. After completing the finetune of flux with Scary Terry I then used kohya_ss gui's lora extractor hidden under utilities to make a 640dim model. This model is way more flexible in my tests than a directly trained lora model. I will also be uploading a 128dim version so that you can see the variation between high and low dim, but the 640dim is recommended.

I started by training a grouping of scary people including Scary Terry for a few thousand steps then dumped everything but the best Scary Terrys. So it also does scary people very well. The people were trained under being a "scary man" or a "scary woman" and all of their skin colors were tagged.

Like my work? Please consider leaving a tip to help fund future LoRAs or commission a custom LoRA for yourself. Higher quality versions of my LoRAs are also available for a monthly donation: https://ko-fi.com/artfulgenie.