After spending nearly a month working with Sophie Molgin I decided to create a new character as a first step to creating a batch of characters that could play parts in stories and scenarios I'm working on.
This is Sophie, an 18 year old ginger woman with grey eyes. She has been created for non-NSFW purposes, but I'm aware that the models I've used to create her can be used for that kind of content.
I'm still getting to grips with good prompt-building and would be very interested to see what others are able to do with these LoRAs.
Re-release: An initial version of this LoRA was posted but had to be removed as I'd included some incorrect images in the dataset and incorrectly tagged a sizeable portion of it, so this is a re-release that will hopefully be more reliable than the initial attempt which was removed due to the state of the training data.
Description
The realistic and cartoon versions of these images were created with the same prompt as the ones to make the SDXL showcase. This was done to try to show how the same prompt based on model can be different.
This has resulted in some settings and backgrounds for the Pony images being incorrect, but from other further testing it all works well enough.
FAQ
Comments (4)
Looks spiffy.
<:)
I'm a bit jealous. Haven't taken the road of trying to create my own artificial people models yet...
Thanks for the compliment.
For what it's worth it's not all that hard. I created a batch of ~1-200 images in a variety of places, outfits and poses, picked about 40 of the best examples that really describe the character I'm looking for then stick it all into one of three different LoRA training tools I'm playing with:
CivitAI (which this model was created with), Invoke AI's training tool (https://invoke-ai.github.io/invoke-training/ because my desktop is powerful enough that it can do this) or a Google Colab document that I found on stable-diffusion-art.com (https://stable-diffusion-art.com/train-lora-sdxl/) which was the first one I used.
If you're interested in learning about how I've done it, I'll be happy to try my hand at documenting my process and post it as an article.
@Nimmog
Never underestimate my ever-fluctuating levels of stupid and laziness.
\;p
I'd certainly read the article, though. Whenever it makes an appearance.
<:)
@NotACompleteIdiot I'll see if I can put that on a todo list and learn how to write articles here (I suspect it'll be pretty easy) and then try to write a coherent guide to document my process.
I think my problem is often short attention span and a lack of time!