Model Overview:
The Neuro-Evolution: I created this Merge for myself, but after checking the permissions of models used in the merge, I decided to post it.
What does the model exactly do and how can it be used?: This model is a checkpoint merge, so it takes the best of merging models (sometimes the worst). the models I merged can be used to create 3d art and anime (mostly anime).
What is the purpose of the Merge: The benefit of Merge is simple – you can find a model that can create good hands, but what if the style of this model is just not for you? The merge can help you.
What method did you use for merging?: I used ComfyUI to merge and test the model.
Is this model LoRA friendly: Yes and no. You can use any LoRA of Character real or fictional, It will create this character with ease, but style LoRAs just give you poop. the best I could do was use pencil-style LoRA.
Did you Bake any VAE: Yes. Blessed2 was baked. At first, I created VAE free model, but after testing, I found that Blessed2 is perfect for this model.
What models did you use for your merge?: the list of merged models is below.
weakened unity v5
CuteHeaven v2.1
Pettanko RouraMashin 2D v2.0
Alladin.3
Prompting:
What are some suggested negative prompt embeddings?: EasyNegative, bad_prompt_version2, badhandv4
What is the recommended way of constructing a prompt for this model?:
(((masterpiece))), (((best quality))) > style > contest
Example: “(((masterpiece))), (((best quality))), anime style, 2d, a woman in a costume with a sword and a moon in the background with a blue sky and stars, Artgerm, rossdraws global illumination, a detailed painting, rococo”
Negative prompt matters – the model loves to generate signatures, so using “(text, watermark, signature, author signing:1.4)” is highly advised. Also, it can generate nsfw pictures, so “nude, naked, and nsfw” should always be in your negative prompt if you want to save time removing nsfw pictures from your PC.
This model likes: ((best quality)), ((masterpiece)) at the beginning of the prompt but maybe it’s just me…
The Settings:
Width: 512
Height: 768
Sampling Eular A
Sampling steps: 30 normally (more than 30 gives you some stupid stuff)
Hires.fix settings:
Upscaler: 4k-UltraSharp
Steps: (10-20),
Denoising: Strength: 0.34 – 0.45
Upscale: 1.5 or 2
Clip Skip: 2
I normally run my model with ADetailer to fix the faces of my created characters, but it is optional. Hires.fix can do the trick.
P.S.: If you like my model and decide to post an image created using it, please make sure it is PG, PG-13, or R max. I understand that I have no right to make any restrictions, but it would be nice of you to do so.