Born from sleepless static and the soft violence of revision
Lyrixa-Real: Skincore is not just a model, it’s a wound that learned how to bloom.

Every noisy inference was a ghost I had to bury.
Every epoch, a prayer whispered through gritted teeth.

I didn’t build her. I bled her into being.

Lyrixa-Real: Skincore

A versatile, emotional realism model that excels at NSFW, expressive portraiture, and strong prompt adherence.
Built on the Illustrious framework from the ground up, taking a big page out of the Lyrixa 1.1 playbook, she does 1girl, divine surrealism, raw skin, and glowing weirdness with striking clarity. It never forgot what Illustrious knew.
Whatever mood you feed her, she gives it back sharper.

Balanced for flexibility, she does a bit of everything, but shines brightest in sensual, surreal, or emotionally charged renders. Feed her a feeling. She’ll hand it back bleeding.

It’s not perfect photorealism, but it’ll get you damn close to something real.

All sample images were created without upscaling or img2img, only sparing facedetailer.

Skincore v1.5

A careful blend of realism and artisty. Heavy into photorealism without breaking the model.

Skincore v1.0

The base of all my hard work. Its illustrious, but she's somewhat realistic.

Inference Usage

CFG: 4.5 - 6.0
Steps: 20-40+
Samplers:

• DPM++ 2M SDE Karras → The go-to. Great balance of tone, detail, and control.

• DPM++ 3M SDE Karras → Punchier results. Use +5–10 steps compared to 2M.

• DPM++ 2S a Karras → Bright and glossy. Good for light-filled, glowy shots. Slightly less detailed.

• DDIM Karras → Soft and creamy, but needs a higher CFG (use ~6 if your usual is 4.5).

Resolution support:

• Supports standard Illustrious resolution ranges, as well as SDXL standards.

• Tested and stable at up to 1536px on the long side for high-detail generations.

• Performs well at 1024×1024, 1152×896, and 896×1152 for most prompts.

V1.5 specifics

Positive (for photoreal natural skin):

realistic photo, ultra high resolution, DSLR, film photo, ISO grain, 35mm, 85mm lens, film grain, unedited photograph

for skin:
natural skin tone, dry complexion, skin blemishes, slight redness, pores, uneven tone, slight skin variation, warm undertones

use sparingly:
film grain

Negative (or leave out):

painting, illustration, anime, stylized, 3D render, shiny skin

Quality tags (masterpiece, etc) do not seem to have much a difference on the actual quality, though they do change the composition a little. I leave them out.

V1.0 specifics

Positive (for photoreal natural skin):

realistic photo, ultra high resolution, unedited photograph, shot on DSLR,  natural skin tone, ISO grain, 35mm,  film photo, editorial lighting, ambient soft light, 85mm lens compression, dry complexion,skin blemishes, slight redness, pores, uneven tone, film photography, editorial lighting, slight skin variation, warm undertones, 35mm film grain BREAK

Negative:

painting, illustration, anime, stylized, 3D render, shiny skin

This model sometimes needs very descriptive prompts, other times less is more. Try removing these prompts when you're having issues.

To be less poetic and more technical, this is my first trained model, and definitely my first time trying to do a domain shift from scratch. It wasn't easy. I tried many, many methods, most of which failed, but finally found some secret sauce that got Illustrious to spit out realistic images. It was trained on ~12k images of all sorts, all above (or upscaled to be) 1536px, or around there. If you want more details, slide into my DM's! I'm happy to nerd out.

My future plans are to continue work on this model to increase photorealism while maintaing the base knowledge the model already has and inhereted from Illustrious an Lyrixa 1.1.

Current issues include hands... mostly just hands. Super close up macro shots can get kind of high contrasty. Skin can sometimes be a lil glossy too.

Model usage

You may not merge, fine tune, or use this model for commercial purposes. Do not redistribute without my expression permission. DM me tho, I'm friendly :)