About this version

DISCLAIMER: This isn’t a trained LoRA - it’s a Schizo-Merge made from my own LoRAs and the sum of differences between ΣΙΗ and NAI VPRED.

After another fail, I think I finally managed to make this LoRA stable enough for use. Unlike the previous version, this time I didn’t make it for ΣΙΗ-2048b but for ΣΙΗ-1.5, fixed the “poisoned” iteration, and tweaked the weights a bit. Everything should be working fine now. I hope. Please. Help.

I tested it on: ΣΙΗ, ΣΙΗ-1.5, ΣΙΗ-2048b, and NoobAI Vpred. It should work fine on these models. However, it doesn’t perform perfect on ΣΙΗ-2048 as I’d like. Maybe I’ll release a separate version for that model later if needed.

And yes, I created this style mostly for myself and my own workflow, so I can’t guarantee it’ll be stable for you. Test it out, hehe.

Anyway, the recommendations are still the same:

1. Weight between 0.5 and 1. I suggest starting at 0.85-0.9 and adjusting up or down. Unlike the previous version, at 0.8-0.9, characters shouldn’t lose their details, but who knows. Test and see.

2. No trigger words needed - just load the LoRA and that’s it.

3. Artist tags are a must. The default style isn’t that strong, so I’m not sure it can stand alone as its own style. It’s more of a direction for a style rather than a style itself.

4. I’ve barely tested compatibility with other LoRAs, so I can’t guarantee stability. Need to test.

5. I made a few comparisons - you can check them here: ΣΙΗ-1.5/ΣΙΗ-2048b and NAI VPRED and ΣΙΗ-2048b

6. Well, Someone told me that this merge doesn’t work on Forge (I’m guessing because of the method I use to mix the LoRAs).

   I tested it on ReForge and ComfyUI, and it should work fine there. I don’t have any other UIs to try it on.