>>43891457>First of all, what's a lora?You can think of them as like hypernetworks but with the power of a dreambooth model. Picrel is an example with a lora of Aegir (azur lane).
>How much have prompts evolved since then?Largely the same but with loras around, trigger words are now more common plus hypernets are now part of the prompt rather than a setting in the settings tab. Loras can also be part of the prompt if you're using the built-in feature.
There is the new Counterfeit model, which has weird prompting though.
>Is NAI still optimal or is there some new kid on the block?We now have new, superior model mixes. Check the rentry. As for base models, AbyssOrangeMix models are now the preferred model of choice. Also new VAEs are being used.
>How the hell do you guys get such extremely high quality images?Using all the new shit we have, essentially. I'm still on my 1060 3gb and my outputs are far superior compared to the NAI days.
>How many of the girls' embeddings have been made already and are they still viable?You can check the thread Mega for embeds but any model that is mixed with hll3.1 can make most chuubas fine without embeds, as long as they aren't an obscure 2view. Nevertheless, you can still use embeds, hypernets and loras to help generate more accurate chuubas if you need to.
Picrel does not have any embed, lora, hypernet or anything extra to help generate Reine. Just a model with hll3.1 mixed in. (Grapefruitjuice pruned)
