How to fix plastic skin in AI-generated images
Plastic skin is a lack of fine texture, usually caused by too much denoising, a high guidance value, face restoration or an upscaler that smooths. Fix it at the source: lower guidance, skip aggressive face restore, regenerate at a higher resolution and add texture with a low-strength second pass. Sensor grain added afterwards helps the overall feel, but it does not paint pores.
Last updated: 2026-10-08
What plastic skin is
The look is familiar: an even, glossy surface with no pores, no fine lines, no small color variation, and highlights that sit on the skin like a coat of varnish. It happens when the model renders the large shapes of a face well and the fine-scale texture badly or not at all.
It shows up for several reasons that stack. Models learn from huge sets of photographs, many of them retouched, so smooth skin is the average. Sampling settings that increase adherence to the prompt tend to push images toward clean, confident output. And the tools applied after generation, such as face restoration and AI upscaling, often smooth detail further in the name of cleaning the face.
Fixing it where it starts
Most of the cure is in the generation settings. The steps above are an order to try them in, one change at a time, so you know which one helped.
Start with guidance. A moderate value usually leaves more texture than a high one. Then look at your face restoration option. These tools are designed to repair distorted faces, and the repair is often a smooth, symmetrical face. If the face is not broken, leave the tool off. Next, look at resolution. Generating a larger image, or a face crop at a larger size, gives the model more pixels for detail than generating small and enlarging.
The texture pass is the most reliable fix when the base image is good. You mask the skin, run image-to-image at low strength and ask for natural skin texture. A low strength keeps the face recognisable and lets the model add fine variation instead of redrawing it. If the face changes, the strength is too high.
What grain can and cannot do
People often reach for film grain. It is worth knowing what it does. Grain is noise spread across the frame with a statistical character. It breaks up the perfectly flat surfaces that make an image feel digital, and it helps different areas look as though one sensor captured them. It does not model skin. A grain layer on a plastic face gives a plastic face with noise on it.
PassReal's grain stage works at that level: it is a sensor simulation, followed by lateral chromatic aberration and a frequency smoothing step, applied to the whole image. Our guide to adding camera noise explains how to do that by hand and what the strength values change. Treat it as the last step, after the skin is as good as you can get it.
Checking your work
Zoom to full size on the cheeks, forehead and nose, where skin texture is easiest to judge. Compare with a real photograph of similar lighting. Look for repeated patterns, which a texture pass can create, and for a visible seam where the masked region meets the rest of the face.
Also check the frequency content. Over-smoothing leaves a characteristic gap in fine detail, which is what the page on frequency artifacts describes. If that gap is still there after your fixes, the image is still smoother than a photograph at that scale, whatever it looks like on a phone screen.
When to stop
There is a point where more passes make things worse. Each image-to-image run changes the pixels, and a face that went through three texture passes starts to show repeated patterns and a loss of likeness. Settle on one good pass and accept that some images are not worth rescuing. Regenerating with a different seed is often faster than repairing a face that was smooth from the first step.
Keep a copy of the version before each step. If a later change is worse, you can return to the previous one without starting again. And judge the final image at the size it will be used. A skin texture that matters at full size on a large print can be irrelevant in a phone thumbnail, and the reverse is also true: a small avatar rewards smooth, clean skin, so decide what the picture is for before you spend time on pores.
Questions
Why do AI portraits look so smooth?
Models trained on retouched photographs learn that polished skin is typical, and sampling settings that favour clean outputs make it worse. Upscalers and face restoration tools often smooth again.
Will adding film grain fix it?
It helps the picture feel less synthetic, but grain is random noise across the whole frame. It does not recreate the structure of pores and fine lines, which has to come from generation or a texture pass.
Does PassReal fix plastic skin?
No. It adds sensor-style grain and chromatic aberration and smooths the frequency spectrum, and none of that is targeted at faces. Fix the skin before you process the image.
Is a higher resolution always better?
Not always. Generating larger can give the model room to place detail, but an upscaler applied afterwards can smooth it again. Compare at full size.
Does fixing skin make an image harder for detectors?
We make no claim about that. Detector scores depend on many properties of a file, and our own tests of processed images against ZeroGPT kept returning 0.96 to 0.97.
Limits
Naturalization operates on the pixels and on the file, not on the content of the image. A subject that is implausible - six fingers, inconsistent reflections, text that does not read - stays implausible after processing, and a human reviewer will notice it. PassReal changes what a statistical classifier measures, not what a person sees.
Results vary by generator, by subject and by detector, and detectors are retrained. No pass rate is published on this page because none has been measured in a way that would still hold next month.
Three images, no card, no expiry.