ODDLY ILLUSTRIOUS v3

This Checkpoint Merge was carefully created through tedious manual dataset curation, and pruning for numerous self trained LoRAs. Using Illustrious XL 2.0 as a base, my dataset focuses and contributes productively to the style diversity and creativity of the model, while giving it some much needed focus.

Can generate up to 2048x2048 resolution
Can do NSFW easily, but training beyond Illustrious XL 2.0 does not include NSFW content.

The default style leans towards a 2.5D anime with strong depth of field.

V3 NOTES

- Anatomy improvements
- Better faces

Really trying to pull this one a little more on the rails with better hands and faces. Compositions may have seen some improvement as well.

V2 NOTES

- Anatomy improvements
- Better compositions
- Better fine details

I will continue to refine this model over the coming weeks. It has come a long way since it's origin model, but still has plenty of room for improvement.

Best Settings:
CFG: 5.0 - 8.0 (Best: 6.0)
SAMPLER: Euler Ancestral
SCHEDULER: sgm_uniform (exponential for latent upscale)
All General Illustrious Prompting Best Practices Apply (booru tags)
1.5mp resolution for best results. (or 1mp and upscale to 2mp)

Enjoy!

Model Info

Download Model