Short answer: 10 to 15 selfies
You need 10 to 15 selfies. That is the sweet spot on almost every AI photo generator, including MakeAiPhotos.
Below 8, your AI photos look generic. Above 20, the extra photos do not really help anymore.
Why 10 to 15? It gives the AI enough variety to learn your real face without overfitting to one angle or one shirt.
Got only 5 selfies right now? You can start, but plan to upload more after seeing the first batch. The likeness jump is huge.
What 8 AI photo generators officially ask for
Every tool quotes a different number. We pulled the current minimums straight from each product page in May 2026.
Notice the spread: 1 selfie at the low end, 12 at the high end. That is not a typo. The reason matters and we explain it in the next section.
Why the numbers vary so much (1 to 20)
Different tools use different AI under the hood. That is the whole story.
Tools like MakeAiPhotos, Aragon, BetterPic, Pica fine-tune a personal AI model on your face. This needs 8 to 15 selfies to lock real likeness.
Tools like HeadshotPro express mode, Canva, and Fotor express use a generic face model with your photo as a reference. They can run on 1 to 3 selfies but the output is less accurate.
Want photos that genuinely look like you across different settings? Use a tool that fine-tunes on your face and give it 10 to 15 photos. Want a fast LinkedIn refresh and care less about per-feature accuracy? A 1 to 3 photo tool works.
The exact selfie mix that works
Counting selfies misses the point. Variety is what makes the AI learn your real face.
10 selfies from arm's length on the same day in the same shirt will underperform 8 selfies across varied angles and lighting. Every time.
Here is the mix we recommend after testing this across hundreds of MakeAiPhotos users in 2026.
What goes wrong with fewer than 8 selfies
Short answer: your AI photos start looking generic or stiff.
Under 8 selfies, the model overfits. It learns one angle, one lighting setup, one expression. Generate a LinkedIn pack from 5 desk selfies and you get usable office shots plus weird outdoor renders.
Under 5 selfies, your AI photos start looking like an attractive person who could be your cousin. Not you.
This is the single most common cause of the 'my AI headshot does not look like me' complaint. The fix is almost never a different tool. It is more varied selfies.
What happens with more than 20 selfies
Short answer: the extra photos barely help and the wait gets longer.
Above 20 selfies, each new photo adds maybe 1 percent realism. Training time grows linearly. 30 photos takes roughly twice as long to train as 15.
Above 30, some tools start flagging your upload for slow queues or QA review.
Exception: if your appearance changed significantly between photos (weight, glasses, beard, hair), uploading 20 to 25 selfies that emphasise the current look helps the model lock the right version of you.
Front camera or rear camera: does it matter?
Mostly front camera is fine. That is how most people shoot themselves.
But mix in 1 or 2 rear-camera shots from 1 metre away. Front camera arm's-length selfies stretch your nose forward and shrink your ears. The AI then learns an exaggerated version of your features.
No rear camera available? Hold the phone further from your face on a stack of books and use the self-timer. Same effect.
When to retrain and what it costs
First batch looks off? Stop generating more photos. Retrain with better selfies instead.
MakeAiPhotos lets you retrain on a fresh selfie set in the same paid session. The realism gain from a single retrain with better input beats generating 50 more frames against a weak model.
Retrain triggers: most outputs look like a generic attractive person, skin tone is consistently off, your hairstyle does not match yours, expressions look stiff in every frame.
All four trace back to input variety, not the AI engine.
Quick guide: how many selfies for your specific use case
Different use cases tolerate different levels of likeness drift. Here is the practical target for each.
Frequently Asked Questions
- How many selfies do you need for AI photos?
- You need 10 to 15 selfies for AI photos in 2026. That is the count where AI photo generators lock in real likeness. Below 8 selfies the output looks generic. Above 20 the gains flatten. Variety in angles, expressions, and lighting matters more than raw count.
- How many photos for an AI headshot generator?
- Upload 10 to 15 photos to an AI headshot generator that trains a personal model on your face. Mix 4 to 6 straight-on, 3 to 4 angled left, 3 to 4 angled right, and 1 to 2 from 1 metre on the rear camera. Single-image tools accept 1 to 3 but produce less accurate likeness.
- Is 5 selfies enough?
- Five is below the recommended floor. Your AI photos will look generic instead of like you. If you only have 5 right now, take another 5 to 10 in soft daylight at different angles before uploading. That extra 30 minutes fixes more issues than any prompt tweak.
- Does uploading 30 selfies give better AI photos than 15?
- No. Above 20 selfies the gains flatten fast. The model has already learned your face by photo 15. The exception is if your look changed significantly (weight, glasses, beard, hair). Then 20 to 25 selfies focused on the current look helps.
- How many selfies does HeadshotPro vs Aragon vs MakeAiPhotos need?
- HeadshotPro accepts 1 to 3 minimum, Aragon asks for 6, and MakeAiPhotos recommends 10 to 15. The lower minimums are designed for fast onboarding but most users see better likeness when they upload closer to 10 to 15 anyway.
- Can I reuse the same selfies on multiple AI photo generators?
- Yes. The same 10 to 15 selfies work on MakeAiPhotos, Aragon, BetterPic, and Pica. Each tool fine-tunes its own model on your uploads so output style differs even with identical inputs. Reusing selfies is a cheap way to test which tool produces the strongest likeness on your face.
- What kind of selfies should I avoid?
- Avoid filters (Snapchat, beauty mode, iPhone Photonic Engine smoothing), sunglasses or hats covering eyes and forehead, group photos cropped tight, burst photos from the same second, very dim photos, and 90-degree profile shots. Filters strip the skin texture the AI needs.
- How many selfies does MakeAiPhotos need?
- MakeAiPhotos recommends 10 to 15 clear daylight selfies. Training runs 15 to 30 minutes, then you get 40 to 80 AI photos across your selected pack for about $15. Below 8 the likeness can drift. Open the generator at /generate to upload.
- Do I need full body shots or just face selfies?
- Just face and head-and-shoulders selfies. MakeAiPhotos learns your face from those uploads and generates new full-body photos in the lifestyle, travel, and modeling packs. You do not need to upload full-body shots unless a specific tool asks for them.
- Can I get good AI photos from 1 selfie?
- Only on single-image express modes (HeadshotPro, Canva, Fotor) and even then likeness is generic. One-selfie AI photos look like an attractive person who shares your features. For photos that recognisably show you, use a generator with 10 to 15 uploads.
- Why do AI headshot tools ask for so many different counts?
- Because tools use different AI architectures. LoRA and DreamBooth-style fine-tuning (MakeAiPhotos, Aragon, BetterPic) need 8 to 15 selfies. Single-image embedding tools (express modes) produce output from 1 to 3 photos but with less accuracy. The minimum reflects the architecture.