Type A1: Vibrant Anime Style - Normal Version
Type A2: Vibrant Anime Style - Kawaii Version
This model is rich in color and lighting and produces very high quality images and it's good for artworks. No detail or quality enhancing loras needed to achieve high quality images. Many loras have been merged to train this model. This model is built on ZenTypeM.
Usage Tips:
-Hires Fix is recommended to fix blurry faces from a distance, 10-16 hires steps and raise overall quality, Upscaler: R-ESRGAN 4x+ Anime6B and denoisening strength 0.35-0.50
-Sampling Method: DPM++ 2M Karras
-Sampling Steps: 30-40 steps
-CFG Scale: 7-10
-Resolution: 512x768 or 768x512
Zen Series:
ZenTypeX: Very Detailed General Purpose Anime.
ZenTypeM: Artistic Anime Style.
ZenTypeA: Vibrant and Colorful Anime Style (Normal and Adorable Versions).
ZenTypeD: Adorable Doll-Like Anime Style.
ZenTypeK: Kawaii Anime Style.
ZenTypeT: Thick Lines Anime Style.
ZenTypeH: Hentai Anime Style.
ZenTypeW: Watercolor Anime Style.
ZenType V: Vintage 90s Anime Style.
ZenTypeB: Eclipse Anime Style.
ZenTypeEX: Experimental Anime Models.
Description
FAQ
Comments (4)
30 steps? I don't think you understand the meaning of Train. Your models are probably just a merged model, not trained.
Both loras and checkpoints are merged into my models, that means it's merged or trained? Sorry i have no knowledge on that.
@Zen12X They're referring to the info on the right side of the model page 'Training'. You entered 'Steps: 30' but it seems you might be referring to the 'sampling steps' when producing an image.
The 'Training' section there is referring to how many steps were used when a dataset was used to train the model with. If this is a merge model like you have outlined, then you wouldn't put anything into this field based on not having trained a dataset as a finetune of a base model.
@Ocean3 I see, i'll keep that in mind thank you.