All models are exclusive to Civitai ! Anyone who publishes my models without my consent will be reported!
Hello everyone!
Trigger Word : NijiEX style
I highly recommend loading my lora models in fp16 if possible
Another new model! 😅
This time I go back to the source with a Niji expressive style model.
The model goes very well with my MJanime model.
Put MJanime at strength 0.5 and the same for N--expressive and you will have nice surprises
Have fun 😉
Description
FAQ
Comments (9)
Seconded on the lora combination mentioned in the description, half this half mjanime. Getting excellent sharp line anime results.
Yes! The best of the models is taken!
The results are excellent !
I'll try to improve this for version 3 of the anime model.
I think flux has a bit of trouble with detailed 2D flat.
Sorry what does it means, "load my lora in fp16" , should the Flux model be fp16, or the Clip ? both ?
This is more for those who use Forge for example.
If no option offers you to load your lora models in fp16, do not worry about it
I use mostly ComfyUI, and Forge when I want to chill. lol also for some reason I get lower quality with Flux in Comfy, no idea why. Anyway
I mean you HIGHLY recommend it but ok let's not worry about it, fine xD I'm tired
oh look what I found, "Cache FP16 weight for LoRA
(Cache fp16 weight when enabling FP8, will increase the quality of LoRA. Use more system ram.)"
No idea that option even existed, this is all so complicate with all the fp8-fp16-nf4bnf64bitAlphaTurboGGUF nonsense
@Tetsuoo Ok I just advise for Forge to use the low bit diffusion option in "automatic (lora fp16)".
This allows LoRa models to be fully supported
@Stan_Katayama Thanks I will try. That option I found sometimes makes a big difference in quality, sometimes no difference at all. I tried to make a few loras for Flux lately but it's still a mystery how to get it right, it's headache material lol
@Tetsuoo Yes. It depends on the training methods of the models.