Sign In

EtherealPerfection2.5D

0

Download

1 variant available

fp16 SafeTensor

Half precision, best balance • 6.46 GB

Verified:

Type

Checkpoint Merge

Stats

277

0

Reviews

Published

Feb 28, 2026

Base Model

Illustrious

Hash

AutoV2
5512B32FAA
default creator card background decoration
Bronze SDXL Badge
MGHerder's Avatar

MGHerder

This is the EtherealPerfection25D checkpoint debut. It is a merge of:

Perfection2.5D (base) by 6tZ

Aurora Fable (second) by jarod2212

And the following LoRAs were baked in: Powerpuff, Smooth Mix, Pony Beautiful Faces slider, detailed further below.

With this pedigree, it excels at drawing structures, anything from a scene containing a living room sofa with windows and plants, to complex airport architecture, while demolishing noise and blips and creating a smooth, realistic output that approaches photorealistic. It's great for Text2Img but Img2Img generation is where I find myself returning to this model constantly as it offers something none of the others I have tried manage to do.

Images in the gallery were generated with the following settings.


EulerA, Simple - native, best approach

CFG 3.5 - 6

Steps: 25-35
OR


DPM++SDE : with Karras - strong results and contrast, careful when using lighting loras

CFG 4.5-6

Steps: 25-35

The following settings function with the model but take longer to calculate, and usually offer small improvements in rendering and smoothness:

exp_heun_2_x0_sde

exponential

20-35 steps

CFG 4-6

When using Img2Img, naturally the desired denoise value can change dramatically. As is common with SDXL models, it tends to redraw the image entirely based off of the prompt at greater than 80% denoise. 70%-77% is the creative zone but can obliterate important details, 50% tends to include most details from the original image. Depending on the style, the model begins to adopt the style of the original image more and more strongly along with all details as denoise decreases below 30%.

This checkpoint uses a Smooth Lighting Enhancer. Applied at .42 strength, it helps smooth out contrasts, adds a built in color/lighting effect, and assists with giving the model space to shade and blend figures without getting confused by harsh contrasts.

This model is good at creating figures, especially beautiful women. It creates semi-realistic faces (2.5D) and can mimic the facial features of LoRAs and lots of famous waifus. Try prompting Tifa face, Aerith face, Yuna face, and so on.

I strongly recommend using ADetailer Workflow (any, try mine) Hands Attention detailer LoRA, and the Pony Beautiful Faces Slider LoRA for best results. You may wish to