Quoted By:
I'm currently doing a thing and I might as well as share here as I did on /h/
picrel is an old pre-2024 sd1.5 gen
I'm trying out SDXL, but alas... No LORA, so...
I have a 710 image big collection of Natsuki DDLC obtained via Hydrus to make up a dataset, which I have pruned out. While I want to train a LORA with it, I also have 2 iq with 1.5 of it dedicated to MMD, so I'm asking if anyone more competent in SDXL character training is willing to take what I've got here and forge an SDXL LORA.
I started the first training attempt, I suspect it will go poorly.
>What did you do in the pruning?
Mostly focused on artstyle appeal and correct body type, so no upsized "assets", and nothing I found too ugly. Some things like some chibis has made it in. I don't want this to be exclusively personal use, so having some examples of other styles seems like a good idea to include.
>Read the-
I have. I have 2 iq. There's also some conflicting information. I can read a lot and I have the patience for it, but I'm not really good without someone to specifically guide me.
>Where's the dataset?
What do you need? Hydrus is capable of outputting the list of tags with each image, which includes omitting some tags. I'm asking how I should sort and (try to) forma to provide a good dataset, at least in usability scale.
>Any requirements you have?
Thanks for being considerate, but as long as you're experienced, I'm sure it'll turn out fine for my daki-making needs.
>I'm not reading all of that
TLDR; 710 Natsuki DDLC images. One 2iq idiot. One wish for an SDXL LORA (Pony, Animagine, doesn't really matter so long as it does good toon daki.). One 3060 12GB.
I have zero fucking clue what I'm doing, if anyone with more experience would also like to see Natsuki DDLC daki, tell me how I should set up the dataset repository, I'll do that, you set a config up and/or train it.
Please help
