This model runs nicely when it is on a weight of 1 in the prompt section. Though, for more variety, putting it at 0.85 gives that. If one is combining the looks with other models, using 0.35 to 0.5 would add a decent mix of the look to other models.
For sampling steps, I like to use 50. A set of 35 steps would give a decent look while retaining the look of the figure. Anything beyond those steps just adds more detail.
A CFG scale of 7 brings amount enough scale for the Lora. A CFG scale of 14 gives a nice dark look to the image. Beyond that, the image would start to gargle a bit. Having a scale beyond that with enough added resources gives a decayed look. A CFG scale of 10 is a nice balance between the standard and dark look.
A standard denosing strength of 0.75 is what I would suggest for images for the character output. If editing one's own pictures with the need to keep their image intact in the image to image area, then a denosing strength of 0.4 with a CFG scale of 21 is what I tend to use.
When using other resources I enjoy using the RevAnimated checkpoint more often than not. Textual Inversions like Easynegative and Deepnegative I tend to use. Badhandv4 and bad_prompt_version2-neg in the negative prompts area lessens the number of unwanted looks well from my experience.
For added fun, using Loras such as the Weregarurumon and Wolf O'Donnell by Cynfall combine well with Zen Aku since they are wolf related. The League of Legends Splash Art textual inversions by Meesh gives a nice anime look to this Lora.
With the new version I added some better quality, and trained it on a different checkpoint. The first was done on the default 1.5 while this new version was trained on a versatile anime one.
1. The rights to reposted models belong to original creators.
2. Original creators should contact SeaArt.AI staff through official channels to claim their models. We are committed to protecting every creator's rights. Click to Claim
