FnFStyleSDXL is focused on humans/characters.

RECOMMENDED SETTINGS:

I have found that using these settings will work best with the FnF Style LoRA:

STEPS - 20+

CFG - 7

SAMPLER - DPM++ 2M Karras

VAE - sdxl_vae

Strength - 0.7 - 1.0

About this LoRA:

Checkpoint Testing:

Testing different Checkpoints has revealed the following to me:

• AutismMix_DPO has the best results with hand accuracy, prompt understanding, and clear results.

• AnythingXL follows the style more closely but hand accuracy is almost near to nothing, prompt understanding is little, and produces messy results.

Styles:

These styles produce different themes in the FnF style that may interest you:

depressing /suicidemouse - produces melancholy and depressing themes for 'Vent' art.

pibbymod/pibby /corruptionstyle /corruption /infected - produces corrupted/glitchy effects combined with white glowing eyes in the prompt.

nsfwstyle/18+ - Produces Explicit content in combination with keywords relative to NSFW.

BODY TYPES:

These body types are if you would want a certain type of appearance with your character:

1girl - Girlfriend figure

1woman - MommyDearest figure

1boy - Boyfriend figure

1man - DaddyDearest figure

Mostly 1girl and 1boy work for female and male characters to produce closer results for the FnF style LoRA.

Study of FNF LoRA:

Comprehension:

Testing the styles shows me that the FNF Style LoRA produces different results if:

• depressing/suicidemouse do work,

• corruptionstyle & pibbymod only work with infected, glitches, corruption, and corrupted added aswell.

• 18+ & nsfwstyle work as expected with NSFW tags added too.

Anatomy & Style:

• DO NOT ADD score_9, score_8_up+ as the FNF Style LoRA will be less effective and worsens hand anatomy.

• Add 3d, realistic, source_anime, source_cartoon, source_realistic to Negative Prompt as a start.

• Lower CFG value changes style to anime

SPECIAL THANKS TO ASTRUSSZZ FOR HELPING ME TRAIN THIS LORA!

Contact on Discord if you have any questions about the LoRA, user is betherz15

描述:

Changes from v1.0 of Friday Night Funkin' Style LoRA:

• Less amount of training images from 1K to 150.

• Different angles than before.

• Focused on subjects in backgrounds for adaptation to different scenes.

Results with the original FNF Style of your prompt.

训练词语: fnfstylev2