Subsurface Diffusion PRO
The red-channel secret behind believable skin
- image
- IMAGE
Here's a fact about real skin that AI models mostly ignore: light penetrates it. Red light has the longest wavelength, so it travels deepest into the epidermis before scattering back out, which is why skin has that soft, warm, slightly diffused quality in the red channel while fine pore detail stays crisp in green and blue. That's subsurface scattering, and it's one of the biggest reasons renders look like plastic - their skin scatters no light at all, so every pore and texture is rendered with harsh, hard-edged precision. Subsurface Diffusion PRO fixes it with a brutally simple trick: blur the red channel, leave green and blue alone.
The cleverness is in the restraint. Blurring all three channels gives you a soft-focus filter, which destroys detail. Blurring only red gives you skin - the warm diffused subsurface layer sits on top of a perfectly sharp green/blue detail layer, and your brain reads it as living tissue instead of painted rubber.
How it works
The node splits the image into RGB channels, applies a Gaussian blur to the red channel at red_scatter_radius, then blends the blurred red with the original red by strength before recombining with the untouched green and blue. That's the whole mechanism, and it's a legitimate, physically-motivated simplification of the subsurface scattering people pay real-time renderers a lot of money to compute. Blue and green - where the sharp pore and texture detail lives - never lose a pixel of sharpness.
The inputs that matter
image- the IMAGE to soften.red_scatter_radius(default 15, 1–100) - how far the red light "scatters" in pixels. 15 is a natural portrait value; tighter for macro, larger for softer skin-light.strength(default 0.8, 0–1) - how much of the diffused red replaces the original. 0.8 is a solid default; back off to ~0.5 if the skin starts to lose its warm blush.
Output: a single IMAGE of the same size.
Installing it
Same pack as the rest - ComfyUI Manager → search "Post Processing PRO", or:
cd ComfyUI/custom_nodes
git clone https://github.com/Guillaume-127/ConfyUI_Post_Processing_PRO
Restart. No requirements, no models - one channel split and one OpenCV blur, instant on CPU.
Where people get tripped up
Two classic mistakes. First, treating the radius like a soft-focus amount - crank it to 60 and you get a red smear over the whole image, because the red channel is now carrying no detail and the color balance shifts warm. Keep it in the 10–25 zone. Second, expecting it to fix skin on its own. Subsurface diffusion addresses why skin looks plastic, but the "too smooth" complaint has a second half - the missing micro-speculars on pores and sweat - and that's Specular Micro-Bloom PRO's job. The two are a matched pair: subsurface gives you the light-penetration softness, micro-bloom gives you the wet surface reflections. Add Film Grain PRO afterward to unify the frame and you've got a portrait that reads as a photo rather than a render.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| red_scatter_radius | INT | 151–100 | — |
| strength | FLOAT | 0.800–1 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |