ZulkiLLuMix


img4 img2 img1 img3

πŸ› οΈ Model Details

Why?

This model is a mix I created by merging several other models that I really liked. The idea was to come up with something unique that could produce images matching my personal taste and creative style. It’s not just about getting good results but also making sure the images feel right and fit the vibe I was aiming for.

At first, this model was made purely for personal use, as I wanted something that fit my specific needs. But over time, I felt it might be worth sharing, so others who come across it can see what’s possible with this kind of approach.

πŸ““ Recipe

This model was created using the following checkpoints:

Merged using ComfyUI with the following method:

model "waiNSFWIllustrious_v80.safetensors" "sdxl" "base"
model "ntrMIXIllustriousXL_xiii.safetensors" "sdxl" "base"
model "miaomiaoHarem_v13.safetensors" "sdxl" "base"
merge "subtract" &1 &2
merge "add_difference" &0 &3 alpha=0.38
merge "clamp" &4 &0 &1 &2 stiffness=0.38

R1 = b - c (subtract operation) R2 = a + (R1 * 0.38) (add_difference with alpha = 0.38) R3 = clamp(R2, a, b, c, stiffness = 0.38)

R4 = Base + R3; with the following weights:

1.0, 0.98, 0.97, 0.90, 0.84, 0.74, 0.68, 0.62, 0.55, 0.50, 0.50,
0.50, 0.42, 0.30, 0.25, 0.18, 0.10, 0.04, 0.02, 0.00, 0.82, 0.50

R4 + ExpressiveH Lora 0.30

Output: ZulkiLLuMix_v1_beta1

πŸ“ Additional Notes

ZulkiLLuMix_v1_beta1_A is using native base model CLIP ZulkiLLuMix_v1_beta1_B is using waiNSFWIllustrious_v80 CLIP without any weight modification

B version somehow has a better result than A version, but it's up to you to decide which one is better.

🎨 Generation Guide

Parameter Recommended Value
Resolution 832x1216 or 896x1152
Sampler Euler a
Steps 25-30
CFG Scale 4-7

Prompt Structure

I don't know if this is necessary, but you can follow this prompt structure. You can experiment with tags order by yourself.

<1girl/1boy/1other/...>, <character>, <series>, <artists>, <special tags>, <general tags>, <other tags>, <quality tags>

Base Quality Tags

masterpiece, best quality, amazing quality, very aesthetic, newest, absurdres

Negative Prompt

bad quality, worst quality, worst detail, sketch, censor, low quality, normal quality, lowres, bad anatomy, bad hands

πŸ“œ License

This model is released under the Fair AI Public License 1.0-SD.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Examples
Unable to determine this model's library. Check the docs .