Hanging-LoRA-V2 for SDXL

Version 2 of my Hanging-LoRA for Stable Diffusion XL (SDXL) significantly extends the functionality compared to version 1. It was trained with more than twice as many main tags (76 instead of 36) and almost 5000 images. It is now possible to create scenes before and after the execution. From standing in front of the noose to standing with the noose around the neck, tightening the noose, to lying on a stretcher after the execution, in the coffin, etc., everything is possible.

In addition, the LoRA was also trained in "crying", but only roughly and no comparison with my extra LoRA. And with hoods, blindfolds and straitjackets.

The LoRA is very neutral in terms of the style used, but has a strong bias towards female victims. Men are possible, but often have to be forced with a higher weighting and possibly negative prompts, which also varies depending on the tags used.

There are over 100 example prompts in a text file so that you don't completely lose track at the beginning. These prompts cover almost all areas of use, were also used to create the promo images, and are intended to serve as a simple starting point.

Most of the promo images were created in Fooocus with Juggernaut8. Especially the anime and other art styles were mainly created with the original checkpoint also in Fooocus. They are unedited except for the last two example pages, where it says so. These were upscaled, touched up with inpainting and in some cases the crying with my Crying-LoRA was added later using inpainting.

As usual with AI and LoRAs (at least for now), the AI does not always deliver perfect images. It depends on the checkpoints used, possibly additional LoRAs and many other factors. In addition, this LoRA was mainly trained for flexibility and not for image quality. So to get really good images, you should edit them. Upscaling, correcting errors in the image using inpainting, also improving faces using inpainting. There are now so many ways to do this and it's worth the extra work.