LimestoneYen is yet another anagram. Anagrams are indeed fun.
This model is essentially a must-have tool for people like me who have not-so-powerful computers at their disposal.
Versions 1 and 2
Recommend using LCM samplers, though I believe it might work with other samplers to varying success. Although no VAE was baked into this model, no VAEs are required (though I recommend the XL_VAE_C_F2 model if you want vibrant colors), neither do you need LCM/Hyper LoRAs; simply set the CFG to 1-2, the steps to 8-12, and you're good to go.
Version 3
This reworked model comes with a VAE baked into it, so now we have vibrant colors. Still testing it out with my LoRAs to see if anything breaks, but so far everything is looking good.
Version 4
Tried baking in an even more vibrant VAE than on version 3, and I tried to finetune a few things to see what stuck and trained with a pruned dataset. Seems to have paid off.
Description
This is an experimental update to ver. 1.0 and the previous iteration of ver. 2.0; the previous 2.0 had terrible quality beyond closeup shots, highly reminiscent of my old LCM models, thus I made some adjustments to rectify that.
This is still a Hyper P model, so you only need 4-8 steps at CFG 1-2 for good results.
If you want colors to be more vibrant, I recommend XL_VAE_C_F2 to be used with it. I could just as well bake it into this model, or make a ver. 2.5 with it, but the uploading process on Civitai is torture enough. F1 also works well, but it can be super saturated.