Selecting the Ideal Input Images

The photos you upload to PhotoGPT AI do more than create one set of headshots.
They train your AI Photographer so it can recreate you in hundreds of presets and custom styles while still looking like you.

💡 Ready to start?
We recommend opening your PhotoGPT dashboard and following the steps mentioned in this guide simultaneously to make the most of it.

How PhotoGPT uses your input photos

When you create a new model in PhotoGPT AI, you upload a small set of photos of yourself.
The AI learns:

• Your face structure and features,
• Your skin tone and hair,
• Your general proportions and style.

Once trained, that same model is used to get unlimited photorealistic portraits of yourself in varied styles and settings. Because all of these features depend on the same model, your input photos are the foundation for every future image.

How many photos should you upload?

In our experience, the sweet spot is:

• 12-15 clear photos of the same person
• Maximum: 20 photos, as long as they are high quality

Aim for:

• A mix of front-facing and slight angle shots,
• A couple of different outfits and locations,
• Similar hair color and style across most photos,
• Uncluttered background (no other people in the background) and,
• No accessories like glasses, hats, etc. unless you want them to appear in every picture you'll create with this model.

More photos are not always better. A small, curated set of strong images beats a big batch of random screenshots and group photos.

✅ Good rule of thumb:
“Would I use this photo to introduce myself on LinkedIn, Instagram, or a portfolio?”
If yes, it’s probably a good candidate.

Model Creation Interface

---

What a good input photo looks like

Listed below are key characteristics of a good input image:

1. Face and framing

• Head-and-shoulders or waist-up shots work best.
• Your face is clearly visible, not cropped at the chin or forehead.
• You’re the main subject, centered or slightly off-center.

2. Lighting and Sharpness

• Soft, even lighting like daylight near a window, is perfect.
• No heavy shadows cutting across your face.
• The photo is sharp, not grainy, pixelated, or motion-blurred.

3. Backgrounds

• Clean, simple backgrounds (plain wall, soft room, outdoors with blur).
• Avoid busy patterns, text, or objects right behind your head.
• No strong color casts from neon lights or RGB setups.

4. Variety that still looks like you

Give the AI a sense of how you look in real life:

• A couple of different outfits (casual, slightly dressed up).
• Slightly varied angles and expressions (neutral, soft smile).
• Outdoor pictures are also fine if the light is good.

But keep the core identity consistent:

• Roughly the same hair length and color,
• No drastic weight changes between photos,
• Similar age (don’t mix very old photos with very recent ones).

Examples of good input images

---

Photos to avoid (common mistakes)

These are the types of images that confuse the AI and usually lead to weaker results:

• Group photos where your face is one of many
• Full-body photos from far away where your face is tiny
• Heavy beauty filters or AR effects (face-smoothing apps, cartoon filters, animal ears, etc.)
• Strong sunglasses, masks, helmets, or big hats that cover your features
• Photos with extreme color filters, neon lighting, or very dark shadows
• Very old photos that no longer match how you look today
• Screenshots of photos from other apps (often compressed and low-res)

If you only have these kinds of pictures, it’s worth taking a few new photos in good light. It will dramatically improve the results you get.

Examples of bad input images:

1. Too many people in the frame

   • Group selfies,
   • Party photos,
   • Travel photos where you’re small in the distance.

   The AI can’t reliably tell which face is “you” to learn from.

   
   Group photos are fun memories, but not good training images as your face is not the clear focus.

2. Unclear or low-quality photos

   • Heavy blur or motion,
   • Very dark or overexposed lighting,
   • Faces covered by hair, hands, or objects.

   These don’t give the AI enough detail to learn how you truly look.

   
   If you struggle to see your own features clearly, the AI will too.

---

Quick tips & Troubleshooting

If your generated images don’t look like you, check for the following:

• Not enough variation
  All your input photos are nearly identical (same outfit, same angle, same room).
  → Replace some photos with slightly different angles and outfits.

• Too many “blocked face” photos
  Many images have sunglasses, hands on the face, or hair covering one eye.
  → Replace them with photos where your whole face is visible.

• Mixed, drastic looks
  Some photos have blonde hair, others dark brown; some with beard, others clean-shaven.
  → Pick the look you want the AI to match and use mostly those photos.

Once your model looks right, you can always refine individual images:

• Use the AI Photo Editor to adjust backgrounds, colors, and small imperfections.
  Learn more in the AI Photo Editor guide.

---

Next steps

You now know how to curate a strong set of reference photos for your AI Photographer.
With the right images:

• Your portraits stay recognizable and flattering,
• Seasonal and creative presets still look believably like you,
• Editing becomes about style, not fixing basic issues.

1. Open your PhotoGPT models tab and upload a curated set of photos.
2. After your model is ready, explore different looks (refer to How to Use Presets guide for quick tips).
3. For more control over wording and styling, refer to our guide on How to Write Prompts.

Your input photos are the foundation. Spend a few extra minutes choosing them well, and your AI Photographer will reward you with realistic, on-brand images in every shoot.