Version: Alpha 0.01
This is a VERY early version, it was only intended for testing only, It's highly likely it'll produce results that are not ideal. I do however appreciate any feedback.
Getting great results with these settings:
Lora Strength: 1.0 CFG: 5.0 (the model's very sensitive to CFG, the lack of training currently is causing it)
Sampler: Euler A
Steps: 50
Recommended Checkpoint: NewrealityX
Prompt guide:
I've been testing ways to ensure the desired look for the LoRA, to help while it has such a low Epoch count. So far I've identified a hand few of keywords that when paired together almost guarantee the desired look and they are:
"Bonnie, anna, highly detailed face, big lips, russian girl, solo, solo teen girl, brown hair"
I got them from digging through barely a quarter of the data set provided to train the LoRA, I'm going to consistently do this over the next few days, so this list will definitely increase but from my experience if you use the resources I've listed and incorporate some of those key words (not all) into your prompt you shouldn't run into any issues with inconsistencies or funky eye renderings. I don't see this being an issue in the future when it has more Epochs in its training, but for now while it's only had 1 epoch with 1 repeat, (in layman's terms this means it's has analysed every photo provided one time only) we'll have to deal with these work arounds.
Here is an example without the keywords provided:
And here's an example of with the keywords:
That is the exact style and face it was trained on, night and day difference to the LoRAs effectiveness. Also please note if you're using FooocusUI you'll have completely different "optimal" settings compared to the ones listed. I have been generating and testing over the past day or so with FooocusUI, and will write a guide specifically for it in the next coming days also.
This model is available on Civitai
