Is there any way to use two loras together(let's say a concept and a character) without distortions in the gens?
I often see some pretty clean results in galleries of concept loras on this website where they also use character LoRAs, so I wanted to know if there's some trick to it.
Hi, if you want to use two or even more LoRAs at once, you need them not to be overbaked or overtrained. This depends only on training "quality" (dataset, training parameters) and nothing more. Some LoRAs looks fried up even on 0.8 - 0.5 emphasis, because some authors are newbies or just don't mind on its quality or compatibility.
For example, i can use ~3 of my LoRAs (character, style, concept) without any issues or artifacts. Or character LoRA + 2 concepts, without any noticeable style change.
You can try to change LoRA Dim to lower values using Supermerger extension, but this may ruin quality a lot, because some layers just being removed in process.