How to make AI photos of yourself

A prompt alone matched our persona's face in 0 of 20 tries. Start from your own photo instead.

Head of Content · Updated November 19, 2026 · How we test
The short answer

Upload one clear, front-facing photo of your face as a persona, then pick that persona and describe only the scene. On PixelPig the photo travels with every request as an identity reference on six image models, so no training is needed. To train a model on your own face instead, on the Pro and Scale plans, use 30 to 40 photos of yourself.

Create your AI model

Upload one clear, front-facing photo of your face as a persona, then pick that persona in the image generator and write only the scene: place, outfit, pose, light. The photo goes with every request as an identity reference, so the model works from your face instead of inventing one. A prompt alone cannot hold a face: in our test on keeping the same face in every AI photo, 20 photos from five models matched our persona 0 times. If you want a model trained on your own face, the next step is a LoRA, and how many photos you need to train an AI model covers that dataset.

The rest of this page is what happens to your photos once you upload them, which photo to pick, and the mistakes that make the result look like a stranger.

What we tested

We traced what PixelPig does with a photo of you: the "Use my own photos" flow through the code that ships today, from the upload screen to the request each model receives, and on to the delete button.

StepWhat the code does
Choose a pathThe first screen offers "Pick a ready-made AI model" or "Use my own photos"
UploadOne face photo, required. "Full body front" and "Full body", both optional. 10 MB per photo
File typesJPG, PNG and WebP. HEIC, HEIF and AVIF are converted to JPEG first
StorageFiles go to a folder under your own account ID, and the database refuses image paths outside that folder
Image generationEvery saved photo is sent as an identity reference, in this order: full body front, face, full body
Face swapOnly the first photo in that order is sent, so a "Full body front" photo is used before the face photo
Models that take a personaNano Banana Pro, Nano Banana 2, GPT Image 2, Seedream 5.0 Pro, Seedream 5.0 Flash, Wan 2.7 Pro
DeleteThe confirmation reads "This permanently removes the persona and all reference photos"

Three results matter for you. No model is trained when you save a persona from photos: the photos ride along as references, which is why one photo is all the form requires. The face swap row means the full-body front photo should show your face as clearly as the face photo does, because face swap uses it first. And the database only accepts photos from your own folder, so another account cannot point a persona at your files.

The other half of the proof is our same-face test: one scene prompt, no reference photo, five image models, 20 photos. The scene was right in all 20. The face matched our persona in 0 of 20. That test did not run the reference route, so it shows what words cannot do, not how close a persona gets.

Pick the right photo of yourself

The upload screen shows its own do and don't list. For the face photo it asks for a clear face, a neutral expression, facing forward, head and chest in frame. Its list of photos to avoid: closed eyes, sunglasses and sharp high or low angles.

In practice, use a recent phone photo taken by a window. Take it at eye level, at arm's length or with the rear camera on a timer, with no filter and no beauty mode. Beauty filters smooth the skin and reshape the jaw, and the model then gets the filtered face as its reference. If you have a tattoo or a beauty spot, the tips ask that it show on the same side in every photo, and a front-camera selfie can flip it.

For "Full body front", stand facing the camera with your face clear and arms and hands in frame. "Full body" can be any angle, as long as your face is visible. Skip photos where you sit or lie down. They are optional, so leave them out rather than add a weak one.

Before you upload, check the photo against this list:

  • Taken at eye level, not from above or below
  • Daylight from a window, face evenly lit
  • Beauty mode and filters off
  • No sunglasses, eyes open, neutral expression
  • Head and chest in frame for the face photo
  • Face visible in both full-body photos, if you add them

Method: your first AI photos of yourself

  1. Sign in on any browser, desktop or phone. PixelPig runs in the browser, so there is nothing to install.
  2. Open Personas, then Create a new persona, choose "Use my own photos" and name the persona.
  3. Upload the face photo. Add the full-body photos only if they are sharp and show your face.
  4. Open the AI image generator, pick one of the six models that take a persona, and select yours.
  5. Write the scene only: "rooftop bar at dusk, black blazer, holding a glass, medium shot, phone photo". Leave out eye colour, hair and age, the persona already holds them. Making AI photos look real has the light and skin words that keep it from looking glossy.
  6. Make four images and keep the best one as your master photo. Compare later batches against it.
  7. If you want a model trained on your face, on Pro or Scale, train a LoRA from 30 to 40 photos of yourself. The training screen asks for the same person in every photo, varied outfits, angles, light and backgrounds, and no other people, sunglasses or heavy filters.

The AI persona page shows the persona screens, and creating an AI influencer step by step covers the same flow for a character that is not you.

Four ways to put your face in an AI photo

RoutePhotos of youWhat holds your facePlan
Prompt onlyNoneNothing. Our test: 0 of 20 faces matchedAny
Reference persona1 required, 2 optionalYour photos, sent with every requestAny active plan
Trained LoRA30 to 40A model trained on your photosPro and Scale
Face swap1, the first in persona orderYour face placed onto an existing imageAny active plan

Most people should start with the reference persona. It needs one photo and works across six image models. A LoRA is the next step when you post often and want more angles than one front photo shows. It runs on its own model in a separate studio on Pro and Scale, not on the six reference models.

A photo of you is still a photo of a real person. Our consent rules say that for photos of only yourself you need no release, but you must be at least 18 and the holder of the account. For anyone else you need a signed release, proof of age from a government photo ID and a record linking them to the persona. PixelPig does not check this on upload, it relies on your confirmation and can ask for the records.

The privacy policy treats face photos as special-category data. It says they are used only to make the output you request, never to identify you, never sold, and never used to train general AI models. Your photos are passed to the AI provider that runs the model you pick, and the policy names those providers. Deleting a persona removes its reference photos, and the policy commits to deleting reference material and trained adapters within 30 days.

Common mistakes

Describing your face in the prompt. The persona already holds the face, so face words add nothing the photo lacks. Let the persona carry the face and give the prompt the scene.

Using a filtered selfie. The filtered face becomes the reference. Use a plain photo in daylight.

Adding a weak full-body photo. Image generation receives every saved photo, and face swap uses the full-body front photo first. If yours is blurry or distant, leave the slot empty.

Expecting one photo to know your profile. A single front-facing photo has no side view. For side angles and wide shots, train a LoRA or keep the camera in front of you.

Posting without a label. If the picture shows a realistic you in a place you never stood, say it is AI. How Instagram detects AI images explains what its label reads and what Meta asks you to disclose, and what C2PA is covers the record some AI files carry.

Bottom line

To make AI photos of yourself that look like you, give the model your face as a photo, not as words. One sharp, front-facing photo in a PixelPig persona is all you need to start on six image models. Train a LoRA on 30 to 40 photos when you want more angles than one front photo shows, and label realistic results when you post them.

Questions

Can I make AI photos of myself for free?
You can save one persona from your own photos before you pick a plan, but generating images on PixelPig needs an active plan. Plan names and what each includes are on the pricing page.
How do I make AI photos of myself on iPhone?
Open PixelPig in Safari or any mobile browser and sign in. There is nothing to install. The upload takes photos from your camera roll and converts HEIC, HEIF and AVIF files to JPEG, up to 10 MB per photo.
How many photos of myself do I need?
One. A face photo is the only required upload for a persona. Two full-body photos are optional. Training a LoRA is a separate step that needs 30 to 40 photos and a Pro or Scale plan.
Is it safe to upload my face to an AI photo generator?
Read the privacy policy first. PixelPig treats face photos as special-category data, uses them only to make the output you ask for, does not sell them and does not use them to train general AI models. Deleting the persona removes its reference photos.
Why don't my AI photos look like me?
If you are not using a persona, the face is coming from words, not from a photo. In our test, one detailed prompt on five image models returned 20 photos and none showed our persona's face. Generate from a persona and fix the reference photo if it is dark, angled or has sunglasses.
Can I make AI photos of my partner or a friend?
Only with their informed consent for this exact use, and you must keep the records. Our consent rules ask for a signed release and proof of age for anyone other than you. For photos of yourself, you only need to be 18 or older and the account holder.
Do I have to say my AI photos are AI on Instagram?
Meta requires disclosure for photorealistic AI video and realistic AI audio, and a realistic AI photo of you can mislead people the same way. We advise disclosing every realistic one.

Create your AI model

Create your AI model

Start here

Read next