Regarding the choice of a fl2va turbo lora
#4
by DaviKar - opened
I see you chose to merge the LightX2V FL2VA Turbo lora into your lora. In my experience this does not work well with the ref2va models.
Would it be possible for to either release a model without the turbo lora merged, or alternatively one with their ref2va turbo lora instead?
Should not add ANY turbo lora for models. They ruin everything.
+1
Including the unmerged base in the repo would make experimentation a bit easier.
How do you use it anyway?
I upscaled latent with 3d upscaler > KSampler with target noise 0.909->0, 3 steps and this LoRA. Just like instruction
But It didn't work
Edit: I used Kjai's lora and it seems he removed turbo lora & lms