>>3023925>Amateur question. If the page doesn't say, how do I know a checkpoint is v-pred or not? For example my understanding is this lora only works with v-pred checkpointsThe original configuration for models was called Epsilon. V-pred is newer and gives models more color depth. The more modern a model is the more likely it is to be V-pred. I believe you can tell for sure by looking at the metadata of a model but I don't know a convenient way to do that.
In any case, LORAs can apply to any model, they just might not have the same effect. So feel free to try those on whatever model you want!
>Are these sort of things placebos? Mandatory? Somewhere in between?There are a million and one LORAs that are supposed to be general quality improvements like those. None of them are mandatory. Pretty much all of them do SOMETHING, but whether they consistently make images better or not is a judgement call. Most likely what they'll do is restrict the potential generations to specific styles and concepts the LORA was trained on, which may or may not be what you want. A decent one might prevent bad gens from looking as bad by limiting how weird it can look, but that might also prevent you from doing more creative concepts and such.
Are you doing local gens? If so experimentation is the best way to figure things out, just try it out and see what you think works best.
