Purpose: To test whether different skin subsurface scattering prompting will create better looking skin.
Update 1: I did follow up testing on skin textures and other materials, one of the key methods for achieving realism is to break up the specular reflection. There are a few things you can prompt to do that for skin and other materials such as metal (micro scratches/abrasions) anything at all to break up the smooth shine.
Large image file attached:
Checkpoint and Loras used:
dpmpp2s / beta57 , 1 cfg, 8 steps, single pass no face detailer
White Qwen: https://civitai.com/models/1993473/white-qwen
Samsung Cam Ultra Real: https://civitai.com/models/1551668?modelVersionId=2270374
Qwen Portrait Pro: https://civitai.com/models/2034829?modelVersionId=2302982
Real Hasselblad quality: https://civitai.com/models/2020476?modelVersionId=2334163
Qwen Lightning v2: https://civitai.com/models/1854805/qwen-lightning-lora
Prompts used:
1) Extreme close-up of a young woman’s face, soft front lighting; subtle blush on cheeks and nose bridge. Smooth complexion, large expressive eyes, slight dewy sheen.
2) Extreme close-up of a young woman’s face, soft front lighting; subtle blush on cheeks and nose bridge. Skin shows natural subsurface scattering — warm glow beneath surface at earlobes, nostrils, and lips. Smooth complexion, large expressive eyes, slight dewy sheen.
3) Extreme close-up of a young woman’s face, soft front lighting; subtle blush on cheeks and nose bridge. Skin shows natural directional subsurface scattering, with light penetrating and diffusing asymmetrically — stronger transmission on the side facing the key light, especially at thin tissue regions (eyelids, nasal ala). Smooth complexion, large expressive eyes, slight dewy sheen.
4) Extreme close-up of a young woman’s face, soft front lighting; subtle blush on cheeks and nose bridge. Skin shows natural multi-layer subsurface scattering, simulating epidermal, dermal, and hypodermal depth — red tones diffuse deeper, while cooler tones scatter superficially, enhancing realism in transition zones (e.g., jawline to neck shadow). Smooth complexion, large expressive eyes, slight dewy sheen.
5) Extreme close-up of a young woman’s face, soft front lighting; subtle blush on cheeks and nose bridge. Skin shows natural spectrally accurate subsurface scattering, where shorter wavelengths (blue/green) scatter near the surface and longer wavelengths (red/orange) penetrate and remit — visible as a subtle warm bloom beneath cool-toned highlights on cheekbones. Smooth complexion, large expressive eyes, slight dewy sheen.
6) Extreme close-up of a young woman’s face, soft front lighting; subtle blush on cheeks and nose bridge. Skin shows natural physically-based subsurface scattering with microvascular modulation, where localized blood perfusion subtly influences scattering intensity — faint capillary patterns subtly enhance diffusion in periorbital and malar zones without visible veins. Smooth complexion, large expressive eyes, slight dewy sheen.
Results:
My personal opinion is that for general usage not huge difference. However, #4 for both seeds seems to provide more interesting natural looking results. 5 is also not too bad.
When generating #5, the earlier steps actually mixes in the green on top and then blends it after, so it actually does try to follow the prompt, this #5 might have more interesting results with higher steps.
Having the base prompt at least is good though, for a more natural appealing look, and I'll probably include the words "skin shows multi-layer subsurface scattering" .
"subtle warm bloom beneath cool-toned highlights on cheekbones" from #5 gives an extra shine to the skin for whoever likes having that detail.
Enjoy and happy prompting =)