Guide · updated August 24, 2026

How to make AI photos of yourself

Making AI photos of yourself is not a filter and it is not an edit of an existing picture. You upload a set of photos, a small model learns what your face looks like from many angles, and from then on it can generate entirely new photographs of you — in a studio you never visited, wearing a jacket you do not own, in light that never fell on you.

The technology is solved. What is not solved is the input, and that is where nearly every disappointing result comes from. This guide is mostly about the twenty minutes you spend choosing photos before you press anything, because those twenty minutes determine whether the output looks like you or like your cousin.

Create my AI model →

The five steps, start to finish

  1. 1

    Gather 10 to 20 varied photos of your face

    Different days, different lighting, different distances and angles. Face clearly visible, no sunglasses, no hats pulled low, no heavy filters, no group shots where the model has to guess which person is you.

  2. 2

    Upload them and start training

    Training builds a personal model from your photos. It runs in the background and takes a while — this is a one-time cost, not something you repeat for every image.

  3. 3

    Pick a pack or write a prompt

    Choose a ready-made pack — professional headshots, dating photos, gym, travel — or describe the scene yourself: setting, clothing, lighting, framing.

  4. 4

    Generate a batch, not one image

    Generate several at once. AI photography is a selection process, not a single shot. Expect to keep roughly one in five and to discard the rest without sentiment.

  5. 5

    Pick the boring ones

    The most dramatic image is rarely the most convincing one. Choose the outputs where the lighting is ordinary and the expression is calm — those are the ones that look like a photograph of you rather than a render.

Which photos to upload (this is 80% of the result)

A personal model learns your face from the examples you give it. If every example was taken in the same room with the same overhead light and the same slight head tilt, the model learns that lighting and that tilt as though they were part of your face. Variety is not a nice-to-have, it is the mechanism.

  • ·Ten to twenty photos is plenty. Fifteen good ones beat fifty mediocre ones every time.
  • ·Vary the lighting: daylight, indoor, overcast, evening. Different light teaches the model what is your face and what is the room.
  • ·Vary the distance: a few close-ups, a few from the chest up, one or two full-body.
  • ·Vary the angle: straight on, three-quarter, slight profile. Do not upload twenty frontal selfies.
  • ·Recent photos only. A model trained on how you looked in 2019 will produce very accurate photos of a person who no longer exists.
  • ·Neutral or mildly smiling expressions. Extreme faces skew the average.

What to leave out

  • ·Sunglasses, masks, or anything covering your face. The model cannot learn what it cannot see.
  • ·Heavy beauty filters or aggressive smoothing. The model learns the filter, then applies it to everything.
  • ·Group photos. Even if you crop yourself out, edge contamination from other faces is a real problem.
  • ·Very low-resolution or motion-blurred shots. Blur teaches blur.
  • ·Photos where you are far away and small in frame.
  • ·Twenty photos taken in one sitting. Same shirt, same light, same day means the model overfits to that day.

How training actually works

Training takes your uploads and produces a small personalised model that sits on top of a general image model. That is why the results can put you anywhere: the base model already knows what a boardroom, a beach and a bicycle look like, and your personal model supplies the one thing it did not know — you.

It is a one-time step. Once your model exists you can generate from it whenever you like, for headshots this month and dating photos next month, without uploading anything again.

Ready to train? Upload your selfies once and generate photos of yourself whenever you need them.

Train your personal model →

Writing prompts that produce believable photos

Once your model exists, the prompt controls everything except your face. Good prompts describe a photograph, not a fantasy: what the setting is, what you are wearing, what the light is doing, and how the shot is framed.

Professional headshot

professional headshot, charcoal blazer over a white shirt, soft window light from the left, neutral grey background, shallow depth of field, shot on 85mm, natural skin texture

Relaxed outdoor portrait

candid portrait sitting at an outdoor cafe table, light knit sweater, late afternoon sun, city street softly blurred behind, natural expression, 50mm, photorealistic

Working environment

environmental portrait at a wooden desk in a bright studio, rolled shirtsleeves, laptop and notebooks in frame, large window light, mid-shot, documentary photography style

Evening, low light

portrait in a dimly lit restaurant, warm tungsten light, dark jacket, shallow focus, film grain, natural expression, photorealistic

Why some results do not look like you

There are only four real causes, and they are all fixable. First, the input set lacked variety, so the model learned a lighting setup rather than a face. Second, the prompt asked for heavy stylisation — dramatic rim lighting, cinematic colour grading, exaggerated poses — all of which trade likeness for drama. Third, you picked the most striking output instead of the most ordinary one. Fourth, the source photos were old, filtered or blurry.

The reliable fix is to generate more and choose more conservatively. Likeness lives in the unremarkable images.

What you can actually use these for

  • ·Professional headshots for LinkedIn, a company page, a CV or a speaker bio.
  • ·Dating profile photos that look like a normal person had a normal weekend.
  • ·Fitness and gym content without hiring someone to follow you around.
  • ·Travel and lifestyle images for a personal brand or social presence.
  • ·Product and brand photography where you are the face of the business.

Eleven role-specific professional packs, from LinkedIn to legal, generated from your personal model.

See the professional headshot packs →

The honest limitations

AI photos of yourself are excellent at volume, variety and cost, and bad at direction. A portrait photographer looks at you and says "drop your shoulder, chin forward, look just past the lens" — that direction is most of what makes a great portrait, and no model does it. What you get instead is a hundred attempts and the freedom to discard ninety-five of them.

They also cannot make you look like someone else without stopping to look like you. If the goal is a photograph people recognise, stay near reality. If the goal is a fantasy image, accept that likeness will soften as the drama rises. Both are legitimate; just decide which one you are making.

Upload once, generate forever

Train a personal AI model from your selfies and generate photos of yourself whenever you need them. From $9.99/month with 200 generations.

Frequently asked questions

How many photos do I need to train an AI model of myself?

Ten to twenty good ones. Beyond that you get diminishing returns, and beyond fifty you mostly add noise. Variety across lighting, distance and angle matters far more than the raw count.

How long does training take?

It runs in the background and finishes in a matter of minutes rather than hours. It is a one-time step — once your model exists you generate from it instantly, whenever you want.

Will the photos actually look like me?

Yes, if your uploads were varied and recent and you choose the realistic outputs over the dramatic ones. The images that look least like you are almost always the ones with the most stylised lighting.

Can I use AI photos of myself on LinkedIn or a dating profile?

Yes, with one rule: it has to still look like you. Nobody objects to an AI headshot, and everybody objects to meeting someone who does not resemble their photos. Pick outputs a friend would recognise instantly.

Is this the same as a face swap or a filter?

No. A filter modifies an existing photo and a face swap pastes a face onto someone else's body. Training a personal model produces genuinely new photographs of you in scenes that never happened, which is why it works for headshots at all.

What happens to my photos and my model?

Your uploads and trained model are private to your account, are not shared with other users, and are removed when you delete your account. Ask any service you upload your face to the same question before you do it.

How much does it cost compared to a photographer?

renza starts at $9.99 a month with 200 generations, against a photographer's session fee for a single sitting. The trade-off is direction: the photographer gets a genuinely good photograph, the model gets you a hundred attempts. Our headshot roundup compares both honestly.

Keep reading