Getting AI Headshots Right with Two Different Eye Colors

Why AI Often Misreads Your Eyes

Think of it like a tired intern scanning a thousand photos an hour. The software looks for patterns, and two different-colored eyes breaks the most basic one. It assumes symmetry. So, it might blend your colors into one muddy average, or worse, pick the dominant eye and copy it. I once saw a result where a client’s striking green eye turned her blue one a weird teal. The AI didn't see "heterochromia;" it saw an error to fix.

Lighting in your reference shots makes this worse. If your left eye is in shadow and your right is in sun, the AI might decide you have two completely different eye *shapes*, not just colors. It gets confused by the contrast, not the iris. Your job is to give it data so clear it can't misunderstand.

Gathering Your Best Reference Photos

You need more than one selfie. Think like a detective building a case for your own face. Provide at least 5-7 photos, but not 20 blurry ones. Each one must serve a purpose.

Start with a straight-on, evenly lit portrait. Phone camera, natural light by a window, no sunglasses. This is your foundation shot. Next, get a photo from each side, showing how light hits each eye separately. Then, include a few with different expressions—a slight smile, neutral. Why? Because your eye shape changes subtly when you smile, and the AI needs to know that contour. Finally, find an old photo where your eye difference is obvious, even if the hair is bad. The AI can pull color data from it.

Here’s the checklist:

• 1 front-facing, neutral expression, even light.
• 1 left-side profile (face turned so right eye is more visible).
• 1 right-side profile.
• 1 with a genuine, soft smile (no squinting).
• 1 close-up of your eyes, if you have it.

Bad lighting is your biggest enemy here. Overhead lights create dark sockets. Flash creates red-eye. The goal is flat, clear, and consistent illumination across all shots.

Writing Clear Instructions for the AI

This is where people are too vague. "I have different colored eyes" is not enough. You must be a technical director. Name the colors specifically. Use common names like "hazel" or "steel blue," but also add a physical descriptor. Where is each eye?

Try a prompt like this: "The subject has complete heterochromia. The left eye is a light, mossy green with a slight amber ring around the pupil. The right eye is a deep slate blue. Please retain this exact difference in the final image." You’ve given it the medical term, the color, the texture, and the location. You’ve told it what to keep.

What if you don't know the precise color names? Pull up a color picker online, find a match, and use that. "Left eye: Pantone 626 C. Right eye: Pantone 7697 C." That kind of specificity cuts through the noise.

The Common Mistake: Over-Correcting in Editing

You get your AI headshot back. The eyes are… close. One looks right, the other is a bit off. Your instinct is to jump into an editor and start tweaking. Stop. This is where you can ruin it.

Over-editing makes eyes look like glass marbles. You slide the saturation up to make the blue pop, but now it looks neon and fake. You dodge the iris to brighten it, but you lose all the natural texture and depth. I watched a client once, in my physical studio, painstakingly clone-stamp her own iris to "perfect" it. The result was a flat, lifeless disc that no longer matched the soulful light in the rest of the photo. The imperfection was her.

Make only one adjustment at a time. Adjust the hue of the off-color eye by 5 points. Look away from the screen for ten seconds, then look back. Does it look more natural, or just more edited? Your benchmark should be your reference photos, not a fantasy of perfection.

Getting a Headshot That Looks Like You

The final test is not on your screen. It’s in the reaction of someone who knows you. Send the image to a close friend with a simple question: “Does this look like me?” Don’t prompt them about the eyes. See if they mention them. If they say, “Yeah, that’s you!” or even, “Your blue eye looks great,” you’ve won. If they pause and say, “Something’s different…” then the AI has averaged you out.

Remember, a professional headshot is a treaty between accuracy and flattery. It should be the best version of you, not a version of someone else. Your distinctive eyes are a feature, not a bug. The goal is to have someone recognize you in a hallway three seconds after seeing your LinkedIn photo.

So, what’s the one thing you should do right now? Find those 5 reference photos. Not tomorrow. Today. The quality of your input is the single biggest lever you have to pull. Clear data in, a recognizable person out. Everything else is just fine-tuning.

If this one does not settle the question, AI Headshots: Wool Peacoat to Plastic Shell? treats the same problem from a different angle.