Free tools Windows power users keep installed
One-click scans. No signup required.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
AI-generated images can feel uncanny when they look recognizably human but combine visual cues that do not quite agree. A face may have realistic skin and lighting yet atypical proportions, eyes rendered at a different level of realism, or details that seem arbitrary. Because the image establishes expectations of a real person, even a small mismatch can become conspicuous.
That mismatch account is useful, but it is not a universal law. Viewers also judge technical quality, fidelity to the subject, whether an image seems prototypical or strange, and what the synthetic depiction means in context. Some modern synthetic faces are highly convincing, so “AI image” is not synonymous with “bad” or “creepy.”
What “uncanny” means in an AI image
Uncanny describes a feeling of unease or wrongness rather than a single visible defect. The image is close enough to a familiar category—usually a human face or body—to activate normal expectations, but one or more cues fail to fit those expectations. The result can be difficult to name: the viewer notices that something is off before identifying why.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesThe uncanny valley is the best-known framework for this reaction. Masahiro Mori proposed that comfort generally rises as a representation becomes more human-like, then may dip when the representation is nearly human but still imperfect. A 2023 synthesis of empirical work cautions that this should be treated as a framework for asking questions, not as a simple, universal curve that predicts every viewer and every image.
#1 Best Overall
Research on computer-generated faces points to a more precise idea: perceptual mismatch. One cue can imply “real person,” while another cue fails to meet the expectation created by the first. This explains why an otherwise polished image can still feel unsettling without containing an obvious glitch.
Why near-human realism makes small mismatches stand out
Human likeness raises the standard an image must meet. In a highly stylized illustration, exaggerated eyes or simplified skin are expected conventions. In a photorealistic portrait, viewers expect the eyes, skin texture, facial proportions, lighting and expression to belong to the same physical world. The more cues that signal a real person, the more noticeable an inconsistency becomes.
| What the viewer sees | Likely expectation | How a mismatch may feel |
|---|---|---|
| Low human likeness, consistent stylization | All features follow the same artistic rules | Usually interpreted as a deliberate style, even when proportions are unusual |
| High human likeness, consistent cues | Eyes, skin, lighting and proportions should agree | Can look natural and convincing |
| High human likeness, inconsistent cues | Photorealism in one element raises expectations for the others | May be experienced as eerie or “off” |
This is why “almost human” is too crude as an explanation. A near-human image is not automatically uncanny, and a more abstract image can also be disturbing if its features look unintended or arbitrary.
The visual mismatches that can create unease
Atypical facial proportions
In a 2009 study, Karl F. MacDorman and colleagues varied facial texture, detail and proportions in computer-generated faces. They reported that atypical proportions were more disturbing on photorealistic faces than on less realistic faces. A face can therefore feel wrong because the spacing or relative size of its features conflicts with the realistic surface treatment, even when no single feature is grotesquely malformed.
Different levels of realism within one face
The same study found that mismatched realism between the eyes and the rest of the face could make a character eerie. For example, highly detailed eyes paired with a flatter or differently lit face create competing signals about what the viewer is looking at. A related 2016 study of inconsistent realism describes this as a perceptual-mismatch account: early cues suggest human likeness, while later cues fail to satisfy the expectation those cues created.
Details that seem arbitrary rather than intentional
MacDorman’s findings also caution against assuming that only photorealistic defects matter. Some abstract faces in the study were eerie because their features appeared unintended or arbitrary. In practical terms, an image may feel unsettling when the viewer cannot infer a coherent design rule behind its eyes, mouth, symmetry or expression.
Familiar examples such as extra fingers
Readers often mention extra fingers, misshapen hands or warped facial features when describing “creepy” AI images. These are real examples of visible inconsistency that some viewers notice, but they are not evidence that every current generator produces them or that they explain all unsettling reactions. A technically clean face can still feel uncanny if its proportions, realism or expression do not cohere.
What the research actually shows
The studies below address different stimuli and questions. Results from modeled faces should not be presented as direct tests of every current text-to-image system.
| Source and date | What was examined | Relevant finding | Important qualification |
|---|---|---|---|
| MacDorman and colleagues, 2009 | Computer-generated faces with varied texture, detail and proportions | Atypical proportions were more disturbing on photorealistic faces; mismatched realism between eyes and face could increase eeriness | The eeriness-rating survey had 458 participants and the human-likeness survey had 407; samples were predominantly young and US-born |
| Study of inconsistent realism, 2016 | How conflicting realism cues are perceived in faces and characters | Cues that imply human likeness can create expectations that other cues fail to meet | This is one account of unease, not a complete theory of every AI image |
| Empirical-research synthesis, 2023 | Evidence concerning the uncanny-valley effect | The concept is useful, but findings do not establish a simple universal law | Stimuli, methods and populations differ across studies |
| Nightingale and Farid, PNAS, February 14, 2022 | 400 StyleGAN2 faces matched with real faces | Selected synthetic faces were nearly indistinguishable from real faces and judged more trustworthy | Faces had uniform backgrounds and no obvious rendering artifacts; the result applies to that sample and task, not every model or genre |
| International Journal of Human-Computer Studies, 2025 | People’s appraisals of generative-AI images | The indexed abstract reports judgments of technical quality and fidelity, along with experiences of images as prototypical or strange; these could raise awareness of social bias and evoke unsettling feelings | The available abstract does not establish the sample size, geography, model, or numerical effects |
Why some AI faces look completely convincing
The strongest counterpoint to stale lists of telltale defects comes from Nightingale and Farid’s 2022 PNAS study. They selected 400 faces synthesized with StyleGAN2, balanced across women and men and diversified by estimated age and race, then paired them with matched real faces. To make discrimination difficult, they used uniform backgrounds and excluded images with obvious rendering artifacts.
For that selected set and task, the authors reported that synthetic faces were nearly indistinguishable from real faces and were judged more trustworthy. They wrote: “Our evaluation of the photorealism of AI-synthesized faces indicates that synthesis engines have passed through the uncanny valley and are capable of creating faces that are indistinguishable—and more trustworthy—than real faces.” The wording matters: this is evidence about the tested StyleGAN2 faces, not proof that all generators, prompts or image categories have crossed a permanent threshold.
Technical quality and fidelity are separate from eeriness
A viewer can dislike an image because it is visibly low quality, because it fails to resemble the requested subject, or because it feels socially troubling. Those judgments may overlap with eeriness but are not identical.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minute- Technical quality: Does the image have coherent edges, lighting, anatomy, texture and detail?
- Fidelity: Does it depict the requested person, object, action or setting accurately enough for the intended use?
- Prototypicality: Does it look like a familiar, expected example of the category?
- Strangeness: Does any feature seem unusual, arbitrary or difficult to interpret?
- Social meaning: Is the image being judged as a harmless synthetic artifact, or as a representation that could reinforce bias or mislead people?
The 2025 HCI evidence is important because it places unsettling feelings alongside technical and social appraisals rather than reducing every reaction to a facial defect. A portrait can be technically polished yet feel disturbing because of what it implies, whom it depicts, or how viewers understand its synthetic origin.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.A practical way to diagnose why an image feels “off”
When a viewer or creator wants to identify the source of unease, inspect the image in this order:
- Estimate human likeness. Is the image meant to be a real person, a stylized character or something in between? The closer it is to a real person, the stronger the expectations it creates.
- Check proportions. Compare the relative size and spacing of the eyes, nose, mouth, jaw and ears. Look for a relationship that appears atypical for the intended style.
- Compare realism across elements. Do the eyes, skin, hair, teeth, clothing and background share the same level of detail and physical lighting? A conspicuously realistic or flat element can break the image’s internal logic.
- Assess technical quality and fidelity separately. Decide whether the problem is a rendering flaw, an inaccurate depiction of the requested subject, or a more general feeling of strangeness.
- Consider context. Ask whether the image is being evaluated as an experiment, entertainment, identity claim, advertisement, news-like depiction or representation of a real social group. Context changes what viewers expect and what consequences they see.
How creators can reduce an uncanny result
No checklist guarantees a natural-looking output, but the evidence suggests practical priorities:
- Choose a deliberate realism level. If a project is stylized, keep stylization consistent instead of adding isolated photorealistic details that raise expectations.
- Audit proportions before polishing. Correct structural relationships first; adding skin texture or dramatic lighting cannot reliably hide an atypical underlying arrangement.
- Match detail and lighting. Inspect eyes, teeth, skin, hair and background for incompatible sharpness, texture or shadows.
- Judge fidelity to the subject. A technically clean image can still fail if it does not represent the requested person, object or action accurately.
- Use human review for consequential images. Have reviewers assess both visual coherence and possible social-bias implications, especially when an image represents identifiable groups or could be mistaken for documentary evidence.
- Test more than one viewer. The 2009 studies used particular participant samples, and reactions vary. A single creator’s comfort level is not a universal measure.
Images, videos and the limits of the evidence
The directly described evidence here concerns still images, especially computer-generated or synthetic faces. A video can introduce additional continuity questions from one frame to the next, but the studies summarized above do not establish numerical claims about current AI video systems. It is therefore safer to apply the mismatch framework cautiously to video rather than assume that findings from still faces transfer unchanged.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Likewise, generator capabilities change quickly. A defect common in an older model may be rare in a newer one, while a new model may introduce different inconsistencies. Claims about what “AI routinely does” should be tied to a named model, date and image set whenever those details matter.
Bottom line
AI-generated images feel uncanny when human-like cues create expectations that the image’s proportions, realism, detail or meaning do not consistently satisfy. The uncanny valley is a useful question, not a universal rule. Research shows both sides of the picture: mismatched facial cues can increase eeriness, while carefully selected synthetic faces can be indistinguishable from—and judged more trustworthy than—real faces. The most reliable diagnosis is to examine human likeness, internal consistency, fidelity, technical quality and social context together.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
