Hungarian tattoo artist Brigi Fuzes admits that about seven out of ten clients now come to the studio with a finished image already drawn by a neural network. Hyper-polished portraits, fantasy faces, animals with cinematic lighting, complex armor, and compositions packed with details—all of this looks flawless on a phone screen. But then the reference must be turned into a tattoo, and that is exactly where the real problems begin.
The perfect picture is a bad sketch
Tattoo artists have always worked from references: photographs, paintings, movie stills, sketches, sculptures, digital collages. Neural networks have added a new source, but with one strange quirk—they know how to make mistakes beautifully. The client sees a pixel-perfect face and thinks the artist’s task is simply to copy the image. In realism, this is almost never the case. Los Angeles-based realist Andrey Kolbasin formulates the question more precisely: how much detail can an image hold when it has to work on a curved, moving surface and remain readable after healing? A screen tolerates infinite detail; skin does not.
Realism relies on things the eye reads instantly: the direction of light, the proportions of facial features, the shape of a hand, the depth of a shadow, the distance between small elements. Of the three types of neural network errors, artists most often encounter three. The first is inconsistent light: a bright spot on the forehead on the left, while the shadow from the nose goes in the opposite direction, plus a “phantom” light source on the jaw that isn’t in the scene. On a digital image, dramatic contrast masks the discrepancy; in a tattoo, light is part of the construction. The second is anatomy: a finger inserted at the wrong angle, an ear lowered by a millimeter, a shoulder that doesn’t connect to the torso, jewelry merging with the skin. The third is the detail itself. A neural network doesn’t care if a tiny pattern will remain readable in a year: it generates eyelashes, pores, strands, fabric texture, and reflections at a scale that only works at high resolution.
Realism is less forgiving
A geometric pattern can absorb a small correction, an ornament can be redrawn around the body, and in illustrative styles, proportions can be exaggerated. Realism does not become more lenient because accuracy is part of the subject. If a portrait is slightly off, the viewer won’t always say why, but they will understand that something is wrong: the distance between the eyes, the shape of the mouth, and the direction of the shadow change the likeness. In an animal portrait, the facial expression relies on a few millimeters around the eyes and muzzle. The body further breaks the composition: an image born as a flat rectangle doesn’t know it’s about to lie on a shoulder, wrap around a forearm, or move with a knee.
At the same time, neural networks are genuinely useful when less is asked of them. Edinburgh-based realist David Corden has described how he uses AI to improve old and blurry photographs, including pictures of deceased relatives that clients want to turn into memorial portraits. Previously, Photoshop was enough to clean up a reference; now, part of the restoration is done faster. But this is a completely different job than asking a generator to invent a tattoo from scratch. A memorial portrait often starts with the only surviving shot: a small print, a faded frame, or a compressed copy from an old phone. Improving contrast, reducing blur, and restoring features give the artist more material. But this requires an eye: AI restoration can invent things that weren’t in the original—the eye will become sharper but slightly change shape, and teeth might appear where the source was too blurry. For memorial work, such changes are critical.
The five-year test and a new consultation format
A fresh tattoo is a deceptively convincing photograph. The blacks are dense, the highlights are sharp, and the transitions are visible in detail. But this very first shot is the worst moment to judge whether all the little things were a good long-term decision. Pigment lives in living tissue: during the healing process and over the years, a tattoo naturally softens. This is not a blowout—a technical failure where pigment immediately spreads into surrounding tissues, which is described in clinical cases. Normal softening is slower and more uniform, and artists factor it into the design in advance. Realism doesn’t turn into a blurry blob after five years, but five years is a design test. Will there be enough negative space between two nearly touching dark zones? Will the face hold up if a few pinpoint highlights become less pronounced? Will the image remain readable from a normal distance if every strand of hair is rendered identically? An experienced artist corrects the image for the future version of the tattoo, not for the snapshot taken on the day of the session.
The consultation has also changed. Previously, it began with an idea, a couple of references, and a conversation about placement. Now, the client arrives with an image that already looks finished, and part of the meeting is spent separating what is useful from what won’t survive the transfer to skin. Before redesigning, they discuss:
- which elements are truly important to the client;
- where the generated image has anatomical or lighting errors;
- which details need more space;
- whether the requested size is large enough;
- what should be the focus and what can disappear;
- how the composition changes for a specific place on the body.
The debate about AI and tattooing is often presented as a conflict between technology and artists, but in the studio, the question is simpler. A neural network can help visualize an idea, assemble a rough composition, clean up an old photo, and provide material that the artist will redesign. Feeling the skin, understanding placement, observing healing, and seeing how the work changes over the years—that is no longer its department. A generated image only needs to live until the next prompt. A tattoo must live through a body. Therefore, the problem is not the tool, but the habit of thinking that an image created without knowledge of healing arrives at the studio already finished. For realists, the work increasingly begins with breaking down the polished digital image into parts and deciding what is worthy of ending up on the skin.