Hello everyone,

I'm finally back (though I was never really gone, just busy). Sorry for not posting anything lately. I’ve been working on my new dual-graphics-card system, and now it’s finally done. Now I can train a LoRA on one GPU while generating images with the other 🎉🎉🎉.

As promised, I’ll start posting the main characters today. I'm starting with Vex Vessar (aka Vex'ahlia Vessar). Let me tell you, it was a pain in the ass — not to get enough data, but to cut the dataset down to around 350 images 🤣😅.
And Vex was only in third place with about 487 images total.

A quick word about the dress outfits: they will most likely be unstable every time, since there wasn’t enough data from that tiny part of Episode 3, and the influence of the main outfit is just too strong. But I decided to include them anyway, because the results are still acceptable — it’s something between the main outfit and the dress, I would say.

Prompt: look at "Trigger Words" on the right side 👉

Changelog: look at "About this Version" on the right side 👉

My Settings (Forge Classic)*:

*note, the example images only use high-res fix.

Checkpoint: waiNSFWIllustrious_v150

Sampler: Euler a

Scheduler: Automatic

Samplings Steps: 30

Hires.fix: R-ERSGAN 4x+Anime6B, 20 Hires steps, Denoising Strength: 0.5

Extension:

Adetailer (highly recommended)