V1
Full model
V1 is more like pastel anime style
V2
Pruned model
Enhances the 2.5D style and sharpens the background
Usage recommendations
I prefer Euler Ancestral or DPM++ 2M Karras
for steps it's better at 20 - 40
CFG scale i think 6 - 8 is best
Resolution 512x768 (Portrait), 768x512 (Landscape)
for VAE you don't need to use it because it's already bakedvae
Prompt
Quality Prompt : masterpiece, best quality, incredibly absurdress, highres, high detail eyes, high detail background
Negative Prompt : (worst quality, low quality:1.4), EasyNegativeV2, ng_deepnegative_v1_75t, bad-hands-5, bad anatomy, interlocked fingers, text, signature
Special Effect : (glowing effect), starry sky, glowing nebula sky, colorful, neon light, fantasy
Of course you don't have to use all the prompts, just take what you need
Hiresfix
use latent (nearest-exact)
hires step you can set as before (i using 35 sampling steps on preview so i use 35 hires steps)
if you use latent (nearest-exact), denoise strength can't be below 0.5 because the image will be damaged (if you force it maybe 0.35 is the limit)
if you copy image preview please first checking hiresfix setting and seed
ADetailer
use face_yolov8s.pt default setting
if you feel the face doesn't match the style try reduce Detection model confidence threshold around 0.2 - 0.3, also you can reduce Inpaint denoising strength if using LoRa or specific face of the character
you can use ADetailer Prompt for example you want 2 different color eyes input "heterochromia, red eyes, blue eyes" or pupil shape, something like that
ADetailer Negative Prompt functions like a regular negative prompt. you can leave it blank
Those are all the settings that I used in the preview