My LoRA was trained on 40 images over 3,000 steps. Pretty straightforward, and it works great with ER_SDE. Res_Multistep, Euler,
Simple, Beta, Bong Tangent.
Minimalistic, and effective. Nothing overly complicated.
I've tested it with several different checkpoints. Some produce slight variations, but overall, it doesn't interfere much with the base model or negatively affect the results.
The sweet spot is between 0.35 and 1. Anything above that, and you start getting some pretty interesting hallucinations.
That's pretty much it. Have fun!
Description
My LoRA was trained on 40 images over 3,000 steps. Pretty straightforward, and it works great with ERSDE.
Simple, minimalistic, and effective. Nothing overly complicated.
I've tested it with several different checkpoints. Some produce slight variations, but overall, it doesn't interfere much with the base model or negatively affect the results.
The sweet spot is between 0.35 and 1. Anything above that, and you start getting some pretty interesting hallucinations.
That's pretty much it. Have fun!