This LoRA can be used to create various cosplayers.
The trigger is: 'A cosplayer photo of'
The dataset features 230 highly curated images built through a massive, multi-model/multi-step Anime2Real workflow. The sources for this WF were high-res cosplayer photography, anime screengrabs, and various artist illustrations. To keep things consistent, I even ran the actual photos through the same pipeline.
Because of this 'standardizing' step, I was able to do much deeper training without 'frying' the results, but keep in mind that it is intended to be a little bit overtrained.
This LoRA is just one stage of a much larger training project I’m currently deep into, but the results were interesting enough as-is that I figured I’d put it out there for you all to play with.
It is NOT intended to be "Maximum Realism" model , so if this is what you are looking for use it in something like the Photoflow WorkFlow (link below).
Enjoy :)