Pony of aussie Josh Duckworth - a beast of a fitness model and uh, content creator .. I wanted to try my new training process on readily available high quality images with some tuned refined image treatment processes I've honed over time. This time I wanted to quantify not only input quality, but output quality on a few factors such as facial matching, CN conformity, prompt adherence and stacked LoRA stability. Since I found a site <a rel="ugc" href="https://www.faceshape.com/face-compare">that </a>measures facial likeness to a fairly impressive degree. I know there others out there that do this, but I found this one to be pretty spot on and fast. Using this to make sure my input data conforms to a high degree of facial likeness, then lo and behold so does the output of the LoRA! One of my quality checks has moved from a subjective me eyeballin' it to now measuring it scientifically to see which epoch I should go with. I even score carded it that I'll use with my future generations (at least ones that I am looking to reproduce someone's likeness). <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d0b54258-0fcc-4d22-8319-09c7102e9855/width=525/d0b54258-0fcc-4d22-8319-09c7102e9855.jpeg" />anywhoo.. 'nuff nerd stuff and back to the main point why you're here: trigger word is joshduckworth enjoy responsibly..

Josh Duckworth - Adult Content Creator (Pony)

Right under CivitAI's 10,000 step limit but with over 300 images in the dataset, it took quite some time to bake but I think it worked out well!

Files with SHA256: 6854cc1d833f3cbaa3931afb48049ee2299b75e50b0245334e62ffbd1500dfa0