PhotoGPTGuides

How to Take Good Selfies for AI Headshots: The Complete Photo Guide (2026)

Bad selfies produce bad AI headshots. Learn how to take good selfies for AI headshot training — the right lighting, angles, expressions, and what to avoid before you upload.

Two-panel hero image showing the same woman on the left as a casual unpolished phone selfie labelled 'your training photo', and on the right as a polished professional AI headshot labelled 'your AI headshot'

You spent 20 minutes taking selfies, uploaded them, and got back someone who looks like you but isn't quite you.

Our team has reviewed thousands of training sets and the conclusion is always the same. The quality of your AI headshots has almost nothing to do with the AI itself. It has everything to do with the photos you feed it. Most people assume the hard part is finding the right tool. It isn't. The hard part is giving that tool something worth working with and that starts with knowing how to take good selfies that are actually built for AI training, not just ones that look nice on Instagram.

AI headshot generators work by finding the "commonalities" across your photos to build a 3D model of your face. Fifteen similar selfies do not give the model fifteen times the information. They give it one data point, repeated fifteen times. The result? An output that is one flat, slightly off version of you.

This guide covers exactly that. Lighting, angles, expressions, what to wear, what quietly kills your results, and a final checklist before you upload. Everything you need to go from a casual phone selfie to a headshot that looks like the one on the right.


Quality Over Quantity: How Many Photos Do You Actually Need?

For PhotoGPT, the sweet spot is 10 to 15 photos in order to train a Personal AI model. That range gives the model enough data to learn your face thoroughly without redundant information slowing the process down. But the number barely matters if every photo in that set is essentially the same shot taken five minutes apart.

If you upload 15 photos taken in the same spot, same afternoon, same expression, same angle and think that they're all different, then trust us, you're making a huge mistake. To the model, they're all the same images. You're treating it like a photo dump whereas the AI needs multiple viewpoints of your face to create your digital twin.

Here's what a strong 15-photo set looks like versus a weak one:

Strong training setWeak training set
Multiple lighting conditionsSame room, same lamp throughout
Straight on, slight left, slight rightEvery photo dead centre
Neutral, smiling, serious expressionsSame expression in every shot
Mix of closer crops and slight pullbacksIdentical framing throughout
Different days or outfitsOne sitting, one outfit

Don't make this process complicated. You're not doing a full photoshoot. A few taken by a window in the morning, a few outside on an overcast day, a few in a different room is more than enough. The goal is not 15 photos. The goal is 15 photos that each tell the model something the others don't.

A practical tip

Don't take all your training selfies in one sitting. Even spacing them across two or three different days naturally introduces the lighting and expression variation the model benefits from, without you having to think too hard about it. The below image shows what a strong training set looks like

iPhone in Space Black showing a camera roll of nine varied selfies of the same woman across different lighting conditions, angles, outfits and backgrounds — captioned 'what a strong training set looks like'

Before anything else get your lighting right first. Everything else builds on top of it.


Lighting: The Single Most Important Variable

Our team insists that if you want to only fix a single thing in your selfies to improve your AI headshots, fix lighting. All the downstream variables that we discussed are influenced by lighting more than anything else. The good news is you do not need studio equipment. All you need is a window and about ten minutes.

Natural window light is your best option

Position yourself so you are facing the window directly, not side-on to it. Facing it wraps soft, even light across your whole face. Side-on creates dramatic shadows that look intentional in editorial photography but give the AI conflicting information about where your features actually sit.

Here is how the most common lighting setups actually perform:

Light sourceWhat it does to your training photos
Window light, facing itSoft and even across your whole face — the gold standard
Overcast daylight outdoorsNatural diffusion, works for every skin tone, zero setup
Direct sunlight outdoorsHarsh shadows the model reads as facial features — avoid
Overhead ceiling lightDownward shadows that teach the model the wrong texture
Ring light, front facingConsistent and clean — good if you have one
Phone flashFlat, blown out, kills surface detail entirely

The best time is mid-morning or mid-afternoon when the light is bright but not harsh. Avoid shooting in direct sunlight streaming through the window — it creates hard shadows under your eyes, nose and chin that the model misreads as texture rather than light.

Three-panel lighting comparison on a dark background showing the same woman: left panel labelled 'window light' with a green check showing soft even light across the face, centre panel labelled 'overhead light' with a red cross showing harsh downward shadows, right panel labelled 'overcast daylight' with a green check showing clean natural diffused light outdoors

The Dark skin tones bias

We understand it as a genuine gap because this is where most selfie guides fall completely silent. There is a clear bias for lighter skin tones. On the other hand, there rarely are any talks on the flat, underexposed results for the medium to deep skin complexions.

If this applies to you, the single most effective fix costs nothing. Take a large white piece of card or paper and hold it below your face, angled upward slightly, to bounce light back toward your chin and cheeks. It fills in the shadows that cause the model to lose detail in deeper tones.

Two-panel comparison on a dark background showing a woman with deep skin tone: left panel labelled 'without bounce card' showing shadowed underexposed chin and cheeks, right panel labelled 'with bounce card' showing the same woman with visibly more even light across the face and filled shadows

Once your lighting is sorted, everything else becomes straightforward. Here is what the rest of your training set needs to get right.


Getting Your Selfies Right: What the AI Actually Needs to See

1. Angles, framing and distance

Getting your angles right is not about finding your best side. It is about giving the model enough viewpoints to build an accurate three dimensional understanding of your face so it can reconstruct you convincingly from any angle later.

You do not need to overthink this. Five distinct positions across your 12 to 15 photo set covers everything the model needs:

Shot typeWhat it teaches the model
Straight on, eyes to cameraFacial symmetry, eye shape, baseline proportions
Slight turn left, 20 to 30 degreesHow your features look from a natural conversational angle
Slight turn right, 20 to 30 degreesCompletes the three dimensional face reference
Chin slightly downJaw definition, how your eyes appear with a subtle downward tilt
Chin slightly upNeck structure, jawline, upper face in context

Hold your camera at arm's length, at eye level. Not above looking down, not below looking up. Your full face should fill most of the frame with a small margin of space around it. Keep your chin and the top of your head both visible. Do not crop tight to save space.

Include a few shots that pull back slightly to show your neck and shoulders. Generated headshots almost always include the upper body and the model handles that transition better when it has reference points beyond just your face alone.

Three-panel image on a dark background showing the same woman demonstrating correct selfie angles: left panel labelled 'straight on' showing full face centred at eye level, centre panel labelled 'slight turn' showing a natural 20-30 degree turn to the left, right panel labelled 'shoulders in frame' showing a slight pullback with neck and shoulders visible — all three panels have green check marks

2. Background: keep it simple, clean and varied

Avoid busy backgrounds. Don't take all selfies with a plain white wall as the background either. The key is to have 3-4 varied plain backgrounds. Just switching the room after every set of 4 selfies is more than enough. What you are teaching the model is that the background is irrelevant and your face is the constant. That is exactly the separation it needs to generate you accurately across any setting later.

3. Expressions: give the AI range, not repetition

Your AI headshots will only ever be as expressive as the range you trained your AI Photographer on. Show it one expression and it learns one expression. Every headshot you generate after that has to work from that single emotional reference point.

If every training photo is the same tight neutral expression, your generated headshots will carry that same energy regardless of the preset or prompt you choose. Corporate headshot, creative portrait, LinkedIn photo will all have the same slightly stiff quality baked in.

Two-panel image on a dark background showing the same woman in identical setup — same outfit, same window light, same background: left panel labelled 'relaxed neutral' showing a relaxed jaw and naturally open eyes, right panel labelled 'genuine smile' showing a warm natural smile — both panels have green check marks


What to Wear and What to Avoid

Your outfit is the second thing the model learns after your face. Most people don't think about this until they see their headshots come back and wonder why every style looks slightly off around the neckline, or why their skin tone reads differently across generations.

The goal is to intentionally keep the outfit boring. By boring, we mean solid colours, simple necklines, nothing that pulls attention away from your face.

Six-panel outfit comparison on a dark background: top row labelled 'what works' showing three green-check panels of the same woman in a white crew neck, navy v-neck, and grey sweatshirt in good light with clean backgrounds; bottom row labelled 'what confuses your personal AI model' showing three red-cross panels of the same woman in a bold patterned jacket, heavy statement earrings in low light, and a group photo with hat and sunglasses

What works

What to wearWhy it helps
Solid neutrals (white, grey, navy, black, beige)No pattern interference as the model focuses on your face
Simple crew neck or v-neckClean edge reference around the neck and shoulders
Fitted clothingAccurate shoulder and body shape for headshot framing

What to avoid

What to avoidWhat actually happens
Bold patterns, stripes, checksPattern bleeds into generated backgrounds and clothing
Logos and graphic textConfuses edge detection around the chest and neck
Saturated colours near your faceCasts colour onto your skin. (The model learns it as your skin tone)
SunglassesThe model can't learn your eye shape, one of its most identity-critical features
Hats and capsObscures your hairline and upper face structure
Heavy jewellery near the faceGets baked into your identity and shows up where it shouldn't
Beauty filtersThe model learns the filtered face, not yours

Your Pre-Upload Checklist

Most training sets fail on one or two small things that are completely fixable before you upload. Run through this before you hit submit.

Your photos

  • I have 12 to 15 photos ready, maximum 20
  • Every photo is a solo shot, no other people in frame
  • All photos are recent and look like how I currently look
  • Hair length and colour is consistent across most photos
  • No very old photos mixed in with recent ones

Lighting

  • At least most photos are taken in good even light
  • No photos taken under harsh overhead ceiling light
  • No photos with strong colour casts from neon or RGB lighting
  • No photos taken in very low light or heavy shadow

Angles and framing

  • I have a mix of straight on, slight left and slight right angles
  • At least one or two photos show my neck and shoulders
  • My full face is visible in every photo, chin and forehead both in frame
  • No tight crops cutting off any part of my face

Expressions

  • I have at least two distinct expressions across my set
  • No heavy squinting or exaggerated expressions in most photos

What I am wearing

  • Most photos have plain solid colour tops, no busy patterns
  • No sunglasses in any photo
  • No hats or caps covering my hairline
  • No heavy jewellery sitting close to my face in most photos

Technical

  • No beauty filters or face smoothing applied to any photo
  • No photos sent through WhatsApp or Messenger before uploading
  • No screenshots of old social media posts
  • Photos are from my camera roll, not compressed or pixelated

Background

  • Backgrounds are plain and uncluttered across my set
  • I have used 3 to 4 different plain backgrounds, not the same wall repeated

Your Work Is Done. Let the AI Do Its Magic Now.

You've done the hard part. Your photos are in, your model is trained, and you now know exactly what separates a strong training set from a weak one. All that's left is to see yourself in a hundred different styles.

Not sure where to start once your model is ready? The Professional AI Headshot Playbook walks you through presets, prompts, and getting the most out of your model from here.

Create your Personal AI Model on PhotoGPT


FAQ

Why does my AI headshot not look like me?

Almost always the training photos. The most common culprits are too many identical photos from the same angle, heavy filters applied before upload, low light causing the model to misread your features, or images compressed via WhatsApp before uploading. Go back through the checklist in this guide and identify which boxes your current set doesn't tick. Retraining with stronger photos will fix it faster than any prompt adjustment.

Is PhotoGPT free to use?

PhotoGPT offers a free tier to get started. Visit the platform to check the latest plans.

Can I use group photos for AI headshot training?

No. The model cannot reliably isolate which face is yours in a group photo. Identity confusion shows up as inconsistent features across your generated headshots. Always use solo photos. If you only have group photos, crop tightly to your face before uploading or take a few fresh selfies.

Do I need professional photography equipment to get good AI headshots?

No. A smartphone and a window is genuinely all you need. The rear camera on a modern iPhone or Android captures more than enough detail for strong AI headshot training. Natural window light is better than most studio setups for this specific purpose.

Last updated on