AI Faces Look Weird? Here's How to Fix Them

You generated a beautiful landscape, a perfect composition, stunning lighting, and then you notice the person in the scene has three nostrils, asymmetric eyes, or that unsettling almost-human-but-not-quite look. Weird AI faces are the most common complaint in AI image generation, and they happen for specific, fixable reasons.

Faces are the hardest thing for AI to get right because humans are wired to detect even the tiniest facial abnormality. A building can be slightly off and nobody notices. A face with one eye 2% larger than the other immediately looks wrong. This guide covers every common AI face problem and the exact prompt changes that fix each one.

Problem 1: Distorted or Melted Faces

This is the most dramatic face failure: eyes in the wrong position, noses that merge with cheeks, mouths that twist sideways. It looks like the face was made of wax and partially melted. This typically happens when the face is too small in the overall image or when too many faces compete for attention.

The Root Cause

When a face occupies less than about 15% of the total image, the AI does not have enough pixel space to render fine facial features accurately. Wide-angle landscape shots with tiny figures almost always produce distorted faces because the AI simply cannot fit proper facial anatomy into that small an area.

Before (distorted)

a woman standing on a mountain peak overlooking a vast valley, epic landscape, wide angle

After (clear face)

close-up portrait of a woman on a mountain peak, valley visible in background, wind in hair, sharp focus on face, golden hour lighting, portrait photography

The fix: Make the face the primary subject. Use portrait framing, close-up, or headshot compositions. If you need a full-body shot, generate it at a higher resolution where the face still gets enough pixels to render properly.

Problem 2: The Uncanny Valley Effect

The face is technically correct, no missing features or distortion, but something feels wrong. The skin is too smooth, too plastic, too perfect. The eyes stare with an empty intensity. It looks like a mannequin or a video game character from 2015 rather than a real person.

The Root Cause

The uncanny valley happens when the AI generates faces that are mathematically perfect but lack the natural imperfections that make real faces look alive. Real skin has pores, slight discoloration, subtle asymmetry, and micro-expressions. Perfect symmetry actually makes faces look artificial.

Before (uncanny)

beautiful woman, perfect face, flawless skin, symmetrical features, stunning

After (natural)

candid portrait of a woman laughing, natural skin texture with freckles, slight laugh lines around eyes, warm window light, 85mm lens, lifestyle photography

The fix: Stop asking for perfection. Add natural texture terms: freckles, pores, laugh lines, sun-kissed skin. Specify natural expressions: laughing, mid-conversation, thoughtful expression. Use candid rather than posed for more lifelike results. Mention specific camera lenses like 85mm or 50mm, which anchor the AI to photographic realism.

Problem 3: Extra or Wrong Number of Fingers

Hands have long been the nemesis of AI image generation. Six fingers, fused fingers, fingers bending the wrong way, or thumbs on the wrong side are all common artifacts. While this has improved dramatically, hands still trip up AI generators more than any other body part.

Before (hand problems likely)

woman waving at camera, full body shot, hands visible

After (hands avoided or controlled)

close-up portrait of a woman, warm smile, hands not visible, soft studio lighting, 85mm portrait lens, chest-up framing

Alternative: controlled hand pose

woman resting chin on one closed fist, thoughtful expression, simple hand pose, portrait photography, sharp focus on face

The fix: The easiest solution is to frame hands out of the shot using portrait or headshot compositions. When hands must appear, describe a specific simple pose: hands clasped, arms crossed, one hand on hip, chin resting on hand. The AI handles defined poses much better than freestyle hand positioning.

Problem 4: Asymmetric or Mismatched Eyes

One eye larger than the other, different iris colors when you did not ask for heterochromia, one eye looking at the camera while the other drifts sideways. Eye problems are immediately noticeable and instantly ruin an otherwise good portrait.

Before (eye issues common)

portrait of a mysterious person, dramatic lighting on half the face

After (consistent eyes)

portrait photograph of a person with deep brown eyes, both eyes looking directly at camera, even studio lighting on face, catchlights visible in both eyes, symmetrical facial lighting, headshot

The fix: Specify eye color explicitly (both eyes same color unless you want heterochromia). Add "looking directly at camera" or "direct eye contact" for consistent eye direction. Use even facial lighting rather than heavy shadow that hides one eye. Add "catchlights in eyes" for lifelike reflections that make eyes look natural.

Problem 5: Multiple Faces Merging

When generating images with two or more people, faces can merge together, share features, or overlap in disturbing ways. One person might end up with part of another person's face, or the AI might generate a surreal composite face between two people standing close together.

Before (face merging risk)

two people standing very close together looking at each other, dramatic lighting

After (distinct faces)

two people standing side by side with clear space between them, one person has short dark hair and the other has long blonde hair, both facing camera, bright even lighting, group portrait photography

The fix: Give each person distinctly different features: different hair colors, different clothing, different heights. Keep physical space between subjects. Use even, bright lighting so both faces are fully visible. Limit group shots to two or three people maximum. Describe their spatial relationship explicitly: "standing side by side" or "one person in foreground, other in background."

Problem 6: Teeth and Mouth Artifacts

AI-generated smiles often produce too many teeth, blurred teeth that blend together, or mouths that look like they contain an impossible number of molars. This is particularly common in wide, open-mouth smiles.

Before (teeth problems)

person with big toothy smile, laughing, mouth wide open

After (natural smile)

person with gentle closed-lip smile, warm expression, relaxed and natural, soft studio lighting, portrait photography

The fix: Use "closed-lip smile," "subtle smile," or "gentle smile" instead of descriptions that force the mouth wide open. If you need visible teeth, use "natural smile showing teeth" rather than "big toothy grin." Avoid extreme expressions that distort facial geometry.

The Portrait Prompt Formula That Works

After testing thousands of portrait prompts, this formula consistently produces natural, undistorted faces:

[framing] portrait of [person description with specific features], [natural expression], [specific lighting], [camera/lens reference], sharp focus on eyes, [one natural skin detail]

Example using the formula:

close-up portrait of an older man with grey beard and deep-set blue eyes, thoughtful half-smile, warm golden hour side lighting, 85mm f/1.8 lens, sharp focus on eyes, natural weathered skin texture

This formula works because it puts the face as the primary subject, adds enough specific detail to prevent the AI from guessing, uses realistic photographic references, and includes natural imperfections that push past the uncanny valley. For more portrait techniques, see our portrait prompts guide and for general prompt improvement, read our prompt formula guide.

Quick Reference: Face Problem Fixes

Frequently Asked Questions

Why do AI-generated faces look weird or distorted?

AI face distortion happens when the prompt does not give the generator enough guidance about facial structure, when the face is too small in the overall composition, or when multiple faces overlap or compete for attention. Faces are the most complex subject for AI to render because humans are extremely sensitive to facial asymmetry and proportional errors that would be invisible in other subjects.

How do I fix extra fingers and hands in AI portraits?

Keep hands out of frame by using tight portrait crops like headshot or close-up portrait. If hands must be visible, describe their position explicitly: hands clasped in front, one hand resting on chin, or arms crossed. Simple, defined hand poses produce much better results than leaving hand positioning to chance.

Why do AI faces look uncanny even when technically correct?

The uncanny valley effect in AI faces usually comes from skin that is too smooth and plastic-looking, eyes that are perfectly symmetrical without natural variation, or lighting that does not match how real faces interact with light. Adding terms like natural skin texture, subtle facial asymmetry, and realistic lighting helps create faces that look genuinely human rather than digitally perfect.

What is the best prompt format for AI portraits?

The most reliable portrait format is: close-up portrait of [person description], [expression], [lighting type], [camera/style reference], sharp focus on eyes. Keeping the face as the primary subject with clear lighting and style instructions produces the most consistent, natural-looking results.

How do I generate multiple people without face distortion?

Multiple faces in one image are challenging because each face competes for rendering attention. Limit group shots to two or three people maximum, describe each person distinctly with different features or clothing, and specify their spatial relationship clearly. For larger groups, consider generating individual portraits and compositing them.