When the family is spread across three cities, the group Christmas photo has to be assembled rather than taken. The method that works on VdoBloom is one person at a time: put each person into the same festive scene, keep their face, and combine the results. Asking a model to insert everybody at once is what produces the floating-heads look people complain about.

Disclosure: VdoBloom is our platform.

Why a one-shot merge goes wrong

A single prompt containing four faces gives the model four problems at once. It has to match four different lighting directions, four camera distances, four focal lengths and four scales, and it has no idea which of those is the reference. So it splits the difference: everyone ends up lit from nowhere, no one casts a shadow, and the heads sit at slightly different sizes against the background. That is the uncanny look, and no amount of "make it realistic" in the prompt fixes it.

Doing it a person at a time removes the guesswork. Each step has exactly one face to keep and one scene to match, so the model has a single job.

The method, step by step

Pick the scene first. Decide the shared setting before you touch anyone's photo — a fireplace, a decorated tree in a living room, a snowy porch. Write it down, because you will use the same wording for every person.
Choose the reference person. Whoever has the clearest, best-lit photo defines the light and the camera distance for the whole group.
Generate each person separately. Open Image Edit, upload that person's photo, and describe the scene in the same words every time, adding where in the frame they sit: left, centre, right, front or back row.
Save every result at the same size. Consistency matters more than perfection at this stage.
Combine them. Assemble the finished people into one image, either by editing the first result and asking for the others to be added one at a time, or by compositing them yourself.
Fix the light last. A single pass that matches warmth and shadow across the whole frame is what makes the group read as one photograph.

If you would rather not use photos of real people at all — for a business card, say — build the scene in the image generator from text instead, or use the AI Christmas Photo presets for a single uploaded person and skip the merge entirely.

Prompt wording for each person

Keep the scene identical and change only the position and the person. For example, the same scene described three times:

First person (left, seated): "Place this person seated on the left of a sofa in a warm living room with a decorated tree behind them, soft window light from the left, keep the face exactly as in the photo."
Second person (centre, standing behind): "Place this person standing behind the sofa in the same warm living room with the decorated tree, soft window light from the left, keep the face exactly as in the photo."
Third person (right, on the arm of the sofa): "Place this person perched on the right arm of the sofa in the same warm living room with the decorated tree, soft window light from the left, keep the face exactly as in the photo."

Repeating "keep the face exactly as in the photo" is not padding. It is the instruction that stops the model from improving someone's features, which is the second most common complaint after floating heads.

Matching light, scale and position

Light direction. Name it in every prompt. If one person's original photo was lit from the right, either flip the scene description for everyone or accept a reshoot-free compromise and fix it in the final pass.
Scale. Decide who is tallest and make the model place the others accordingly. A grandmother rendered at her son's shoulder height is the single most obvious giveaway.
Contact. People should touch something. Hands on the back of a sofa, a shoulder against a doorframe, feet flat on the floor. Figures that touch nothing look pasted in.
Depth. Two rows beat one. A standing row behind a seated row gives the scene somewhere for the eye to travel.

When the family is in different countries

This is the request behind most of the search traffic for merging Christmas photos, and it is the case the method suits best. Decorations differ, seasons differ, and someone is usually in a warm climate while someone else is in snow. Settle on one setting in the prompt — usually a northern winter scene, because that is what the model draws most convincingly — and describe everyone in the same clothing weight: coats and jumpers, not a T-shirt on one person and a parka on another. If the distance is the point of the card, make it explicit in the wording you add afterwards: "together at last" or the names of the cities reads warmer than pretending everyone lives in one house.

What to do when it still looks off

Work on one problem at a time, and check in this order: light direction, then scale, then contact, then skin tone, then the background. If a face has drifted from the original, regenerate that person alone rather than regenerating the whole group. If one person's clothes fight the scene, change the clothing description, not the person. And if the group still refuses to sit together convincingly, reduce the number of people in the frame and put the rest in a second photo — two clean half-groups read better than one crowded failed merge.

Only use photos people have agreed to

Uploading someone's photo to generate their likeness requires their permission. Unconsented photos of real people are refused. In practice this is simple for a family card — ask before you use the picture — but it rules out pulling faces from an old group shot to fill a gap, which is worth knowing before you try it.

Preparing each photo before you upload

A merged photo is only as good as the worst picture in it, and a few minutes of prep removes most of the mismatch before anything is generated.

Match the camera height. Photos taken at eye level merge far more easily than a mixture of a low angle and a shot from above, because the model otherwise has to reconcile two perspectives as well as two lights.
Prefer front-facing. A three-quarter turn is workable; a profile is not, because the model has to invent the rest of the face to turn it towards the camera and that invention is where the likeness goes.
No sunglasses, no hat brim. Anything covering part of a face leaves the model guessing.
Pick photos of similar contrast. A picture shot in bright daylight sits badly next to one from a dim room, even when both are correctly exposed.
Scan old prints flat. A photo of a photo on a table brings the table, the glare and the angle into the upload. A flat scan gives the model something clean to work from.

Do the prep first and keep the originals untouched. When a merge goes wrong it is usually one bad source photo rather than a bad prompt, so it is worth knowing which pictures were clean before you start changing wording.

If the group refuses to sit together

Sometimes the assembled picture keeps fighting you, and the cause is almost always sources that differ too much in light or angle to reconcile. The fallback that always works is to stop combining people in one scene and lay out the portraits instead: one image per person, generated with identical scene and light wording so they look related, then arranged as a row or a grid with a band underneath for the names and the year. Four people, each unmistakably themselves, read better on a card than one frame where two of the four have drifted — and a grid is far quicker to approve than a merge you keep regenerating.

Questions people ask about this

How do I make a group Christmas photo when the family is in different places?

Settle on one scene, then edit each person into it separately using the same scene wording, and combine the results. Keeping the lighting description identical across every person matters more than any other single step, because it is what makes the assembled photo read as one picture.

Why do my merged family photos look like floating heads?

Because everyone was generated in one pass. With several faces in a single prompt the model cannot decide which lighting and scale to follow, so the heads end up at slightly different sizes with no shadows. Generating one person at a time and matching light and scale fixes it.

How many photos should I merge into one Christmas card?

Groups of three to five people hold up best. Beyond that, scale errors multiply and the result gets busy. If the family is larger, make two half-groups and use them as a pair, or make one image per household and lay them out as a grid.

Can I merge photos of people who have died into a Christmas photo?

You can if the family agrees, and it is a common reason people learn this technique. Treat it as a memorial image rather than a candid, tell the relatives who will see it, and be careful with the wording — a cheerful "look who is here" over an image including someone who has died lands badly.

Do I need editing software to combine the images?

No. Editing each person into the shared scene inside Image Edit does most of the work, and the final light-matching pass can be done there too, by asking for consistent warmth and shadow across the frame.

Will the faces still look like my family afterwards?

Yes, provided you upload clear, evenly lit photos and say "keep the face exactly as in the photo" in each prompt. Faces are kept from the upload, so the card shows your family rather than a described impression of them.

Can I merge two photos into one Christmas picture?

Yes, and the same method covers two people or six. For two, generate the first person into the scene, then use that result as the starting image and ask for the second person to be added beside them, so there is a real reference for the light and the scale. Asking for both in one prompt is where floating heads come from.

What if one person's photo is old or low quality?

Generate that person first and on their own, then check the face closely before anyone else joins the frame. A soft, faded or grainy source gives the model less to keep, so expect a looser likeness and have a second photo ready. If no better picture exists, keep that person smaller in the frame and close to the reference person, who sets the light for everyone else.

→ Edit a photo into a Christmas scene · Make an AI Christmas photo · Create a free account

Christmas Family Photo From Separate Photos: One Frame, Any Distance