How to Add a Person to a Photo (Tested, With the Two Tells)

Jul 29, 2026

Someone is missing from the photo — a grandparent who could not travel, the friend who took it instead of standing in it. You can close that gap now without owning Photoshop. Below is the method, tested on a real photo, including the two places it goes wrong.

The short answer: how to add a person to a photo

You need two images — the scene photo and a photo of the person — plus one instruction that says where they stand, how big they are, and where their feet meet the ground. Any image model that accepts two inputs at once can composite them in a single pass; we ran the example on this page on Nano Banana 2, which kept the added person's face and clothes intact.

The edge of the cutout is no longer what gives an edited photo away. In our run the two tells were depth of field and contact shadow: the inserted person came back sharper than the rest of the photo at her distance, and the shadow under her shoes was faint until asked for.

Two source photos and the composed result: a woman from a studio photo inserted into a wet city street photo, with her sweater, trousers and loafers unchanged

Why the old way to add a person to a photo no longer applies

The older guides teach the same sequence: cut the person out, feather the edge, paste onto a layer, colour-match, hand-paint a shadow. That existed because software could not understand a photo — only pixels and selections, so a human supplied the understanding.

Image models supply it now. Given a scene photo and a person photo, the model already knows which pixels are the person, what a pavement is, and how large a human should look twenty feet down a street. You are no longer operating a cutting tool — you are directing something that can already see the photo, which moves the skill from selection technique to saying what you want.

How to add a person to a photo, step by step

1. Pick the scene photo and read its light

Open the photo you want the person added to and check one thing: where the light comes from and how hard it is. Overcast daylight is the easiest case in any photo — soft, directionless, so anything you add blends. Hard afternoon sun is the hardest: every object in the photo agrees on a shadow direction, and a newcomer who disagrees is instantly wrong.

Given a choice of scene photos, pick the flatly-lit photo. That single choice does more than any prompt wording.

2. Pick the person photo — full length beats a headshot

The person photo should show as much of the body as the final photo needs. If they will be standing, use a photo where they are standing, ideally head to shoes. Asking a model to invent legs for a chest-up photo is far less reliable than giving it legs to copy, and invented legs are where proportions go strange.

Front-on or three-quarter both work. What matters more is that their light is not fighting the scene: flat studio light drops into an overcast street easily, while hard sunlight brings that key light with it.

3. Write the instruction like a stage direction

This separates a believable photo from an obvious one, and it is where nearly every tutorial goes vague. Do not write "add the woman to the photo". Write where she stands, what must stay unchanged, and how she meets the ground.

Here is the exact instruction used for the photo on this page:

Add the woman from the second image into the street scene of the first image. Place her standing on the pavement a few steps behind and to the right of the walking woman, at a size and perspective that matches the pavement she is standing on. Keep her face, hair, cream knit sweater, dark trousers and brown loafers exactly as they are in the second image. Match the overcast daylight of the street scene, add a soft contact shadow under her shoes on the wet pavement, and keep everything else in the first image unchanged.

Four things in that paragraph are doing the work:

  • A position relative to something already in the photo — the model can see the walking woman, it cannot see your mental coordinates.
  • A scale anchor ("matches the pavement she is standing on"), which prevents the giant-or-doll error.
  • An identity lock — face, hair and each garment named. Unnamed details drift.
  • A ground contact instruction, or you get a person hovering above the pavement.

4. Judge the result on three specific things

Do not ask whether the photo "looks real". Ask three narrow questions, in this order:

  1. Are the feet planted? Look for a shadow or reflection where the shoes meet the ground. The most common tell by far.
  2. Is the sharpness right for the distance? If the trees behind the person are soft and the person is razor-sharp, the eye reads paste-in without knowing why.
  3. Does the light agree? Which side of the face is brighter, and does that match the rest of the photo?

If one is wrong, change that one thing and run again. Rewriting everything when only the shadow was off is how people burn twenty attempts.

What our own test photo actually shows

The composed photo at full size: the inserted woman standing on the wet pavement behind the walking woman, both in the same photo

The photo above was made in one pass, no retouching afterwards. What carried over correctly: her face, her hairline, the ribbed collar of the cream sweater, the dark trousers, the brown loafers. Her scale against the pavement slabs is right, and she is lit as overcast daylight rather than as the studio she was photographed in.

What is not perfect, said plainly: she is slightly too sharp for her distance. The trees and buildings at her depth are softened by the lens and she is not, so a careful eye reads her as belonging to a different photo. Her contact shadow is there but light — on wet stone you would expect more reflection under the shoes.

Both are fixable in a second pass by naming them ("soften her to match the depth of field at that distance; strengthen the reflection under her shoes"). Worth stating because no competing guide mentions either, and they are what a viewer notices first.

The honest limits when you add a person to a photo

A face you know is judged far harder. A stranger added to a street scene passes easily. Your own mother, added to a family photo, is examined by people who know her face precisely — a small change to her smile reads as wrong even in a technically clean result. Use the sharpest front-on photo you have, and expect several runs.

Group photos are harder than a single insertion. Each extra person adds occlusion questions — who overlaps whom, whose arm sits in front. Add one at a time, feeding each finished photo back in, rather than asking for three at once.

Hands and joins are the weak points. Standing near someone is far more reliable than touching them.

Nothing recovers detail that was never captured. If the person image is small, blurry or badly lit, the result inherits all of it. A model can relight and rescale; it cannot invent what the camera never recorded.

Where to add a person to a photo: phone, desktop or browser?

The searches split by device — iPhone, Android, Canva, CapCut — but that split matters less than it used to, because the work happens on a server either way. What actually differs:

  • On a phone, editing apps such as CapCut and Picsart ship their own AI insert features, driven by taps rather than a written instruction. The trade is being limited to what that app offers.
  • In a browser, you keep the full instruction — position, scale, identity lock, shadow — and can run a second pass at the one thing that was wrong.
  • In Photoshop, most control and most skill: worth it for professional compositing, overkill for one family photo.

You can add a person to a photo in the browser on Kavel: upload the scene photo, upload the person photo, and paste an instruction written like the one above. The example here used the same model and the same two-image instruction, submitted through the API rather than the browser form. To generate a person instead of inserting a real one, the full body AI generator builds a standing figure from a description.

FAQ

How do I add a person to an existing photo?
Upload two images to an AI photo editor that accepts multiple inputs — the scene photo first, the person photo second — then write an instruction naming where the person stands relative to something already in the photo, what to keep unchanged, and to add a contact shadow under their feet. One pass, under a minute.

How do I add a person to a photo on my iPhone?
The Photos app cannot do it. Use an app with an AI editing feature (CapCut, Picsart and similar) or a browser-based editor, which behaves identically on iOS because the processing happens on a server. On a phone the limit is the interface, not the phone.

Can I add a person to a photo for free?
Yes, for a small number of photos. Most tools give free credits then charge — Kavel included. Watch for tools that are free but watermark the photo, which defeats the point if you plan to print it.

Will people be able to tell the photo was edited?
Check three things and so will they: whether the feet cast a shadow, whether the person's sharpness matches the photo at their distance, and whether the light hits them from the same side as everyone else. Get those right and a casual viewer will not notice.

Can I add a person who has died to a family photo?
Yes, and it is one of the most common reasons people search for this. Use the clearest front-on photo you have of them, add them one at a time rather than into a crowd, and expect several attempts before the face reads right. The result is a keepsake, not a document; treat it that way.

What resolution should the source photos be?
As high as you have. The result is limited by whichever source is worse, so a 4000-pixel scene and a 500-pixel person produce a soft, obviously-inserted figure. If the person image is small, crop it as little as you can.


Last updated: July 2026. The photo on this page was generated on Kavel with Nano Banana 2 from two source photos, in one pass, with no retouching.

Want to make your own AI video?

Turn an idea into a Kavel video in seconds. Pay only for what you use.