Deryn Sharp (Leviathan)

A lora trained on a dataset I generated using Qwen Image Edit to create various poses and expressions from an original image I created. Trained using Nova Anime V9 illustrious. It seems to want to generate her at the wrong proportions at 1.0 strength, I think that's either a result of my data set or the checkpoint I used.

Original source image I made is:

All data set images were on a white background, I removed all background tags so hopefully it will accept any background without problems. My first lora so if anyone has suggestions to improve it I'm happy for any advice. I used a ComfyUI workflow and Prodigy at 1200 steps. It works a bit better with the carabiner detail especially with a second upscale sampler.