I often find the concepts I tried to remove will appear when applying negative weights. Say that images of a man with heavy beard are usually blurry and wearing hat, I needed to deal with that by balancing the features so most things in the images don't change too much. I might have remove blurry, hat, etc in positive side, then images can be blurry or wear a hat when you apply negative weights. But this depends on the concepts, if the base model is fully versatile on the extreme concept and doesn't generate too many side effects, it should work fine as a normal double sided slider. But what would you want to do with negative weights anyway, most characters on the site don't have beard.

The only thing that is saying is just the models you often use are very different from the ones I use. You simply cannot make sure every model to behavior the same way on every checkpoint, especially on the ones with different styles. Their Unets are different. If you need one to work 100% on the model you use most often, you can consider giving me a commission request. I will then happily train the model on the one you are using. btw I am sure you can easily fix that by tweaking the setting of the LoRA.

@victorc25744 imho base model is not the best model to train sliders as some concepts are not well represented. and the results are probably going to be less desired for me as well.

what do you mean?? That is an enough amount of beard, no?

@ki_pah206 I am pretty sure you don't even need lora if you mask it, but you can try put this lora in negative weight, that should work,

@ki_pah206 thank you, I would like to know more details, but I don't want to put email in public, do you have discord? My username is the same overthere, you can send me a PM.

@ki_pah206 yes, I am on civitai server and here's my discord ID 1116651608554606712 :)