You can usually tell an AI Christmas photo within a second, even if you cannot say why. The giveaways are small and specific: hands with the wrong number of fingers, teeth that merge into one another, jewellery that fuses to skin, text in the background that reads as nonsense, and a light source that appears from nowhere. Knowing those tells does two things — it lets you spot other people's, and it tells you exactly what to fix in your own before you send it.

Disclosure: VdoBloom is our platform.

The tells, in the order they appear

Hands. The classic. Fingers that bend the wrong way, a thumb on the wrong side, six fingers, or a hand that disappears into a sleeve that has no opening. Look at hands first, always.
Teeth. A smiling mouth with a continuous white band instead of individual teeth, or a number of teeth that does not fit the face.
Eyes. Reflections that do not match each other, pupils of different sizes, or a gaze that lands slightly past the camera.
Jewellery and glasses. Earrings that merge into an ear, an earring missing from one side, spectacle arms that vanish into the temple, a necklace that becomes part of the neck.
Background text. A card, book, poster or tag with lettering that is nearly a word. This is the most reliable tell of all, because models draw text as art rather than type.
Repeated patterns. Identical baubles in a neat row, a knitted jumper where the pattern forgets itself halfway across, wallpaper that repeats at the wrong scale.
Light that does not come from anywhere. Faces lit evenly from all directions, or a strong glow with no visible lamp, window, fire or tree to explain it.
Physics. A mug with two handles, a mug handle that nobody is holding while the hand is elsewhere, a chair leg that does not reach the floor, a dog with a tail emerging from the wrong place.
Skin and hair. Uniformly smooth skin with no pores, hair that turns into fabric at the shoulders, or a beard that has no edge against the jaw.
The whole frame. Everyone slightly too symmetrical, the room a bit too tidy, the snow too evenly distributed.

Why AI Christmas cards bother people

The objection is rarely about the technology. It is about what a family photo is for. A Christmas card that shows a family together in a scene they were never in is a small claim about a moment that did not happen, and when the card is signed by hand the mismatch is what irritates people. There is a second, more practical concern: a photo card is also a statement about how the family is doing, and a generated image can imply a togetherness or a house or a holiday that is not real. That is why the backlash lands hardest on cards that pretend to be candid photographs rather than on obviously illustrated or stylised ones.

The same objection does not apply to every use. A festive portrait that is clearly a portrait, a thank-you image, a business greeting, or a stylised illustration built from a real photo reads differently, because nobody is being asked to believe it is a snapshot.

Doing it in a way you are happy to send

Keep the likeness honest. Say "keep the face exactly as in the photo" when you generate, and check that it did. Subtle beautification is the thing people notice most, because they know the face.
Choose a scene that is obviously a scene. A studio-style portrait, a painted finish or an illustrated card reads as what it is. A fake candid snapshot in a fake kitchen does not.
Say so if it matters. For a card that goes to family and friends, a short line — or a caption on the post — removes the whole problem and costs nothing.
Do not put anyone in who was not there. This is the line. Adding a relative who has moved away, with their agreement and their real photo, is a different act from inventing a scene and signing it.
Get agreement for every face. You confirm permission for each photo before generating, and unconsented photos of real people are refused. That rule exists to protect exactly the people these images depict.

Checking your own image before you send it

Look at the hands first, both of them, at full size.
Zoom into the eyes and check the light reflections match.
Read every piece of background text. If any of it is nearly a word, regenerate.
Check jewellery, glasses and collars for anything merged or missing.
Ask where the light is coming from, and whether a shadow agrees with it.
Look at anything the subject is holding, and check the object makes sense.

If you are generating this yourself, the AI Christmas Photo tool keeps the face from your upload, which is the fastest way to avoid the beautified-stranger problem. If you are checking an image someone sent you, remember that the reverse is also true: a convincing face is not evidence that the rest of the picture is real, so check the background and the hands instead.

Tells that are not tells

Some things that look suspicious are not evidence at all, and treating them as proof is how people end up accusing each other wrongly.

Film grain and softness. Deliberate vintage processing and a genuine old print look similar, because both lose detail in the same places.
Uneven or unflattering light. Real indoor photographs are full of it. Generated images tend towards clean, even light, so harsh light is mild evidence of a real photograph rather than the reverse.
A blurred background. Portrait mode on a phone produces the same tell people point to in generated images, including the soft edge that cuts into hair.
A tidy room. People do tidy up before a family photograph is taken.
A soft, low-resolution file. Compression hides detail in exactly the way a generated image does, so a small or grainy picture proves nothing either way.

Where a generated photo is fine, and where it is not

The same image can be unremarkable in one place and irritating in another, and the difference is almost always what it is being presented as.

Fine: a festive portrait, a stylised illustration, a thank-you card, a small business greeting that obviously reads as a design, and a memorial image made with the family's agreement. Nobody is being asked to believe these are snapshots.
Fine with a note: a family card where everyone agreed to the photo and the scene is understood as a scene. One short line removes the whole question.
Not fine: a generated image presented as documentary evidence of a gathering that never happened, a likeness used without permission, and a picture of someone else's child made for a public post.

If you are the one sending it, the test is simple: could you answer honestly if the person receiving it asked whether it was real?

Questions people ask about this

How can I tell if a Christmas photo is AI?

Look at the hands, then the teeth, then any text in the background. Fingers are the most common failure, merged teeth are the second, and background lettering that is almost a word is the most reliable tell of all. If all three are perfect, check jewellery and the light.

Are AI Christmas cards a problem?

Generating a festive image is not the problem. Presenting a constructed scene as a candid photograph of a moment that did not happen is what people object to, especially when it implies the family was together or somewhere they were not. Making it clearly a portrait or an illustration avoids the issue.

Should I admit my Christmas photo was made with AI?

For a card to family, saying so is the safer choice and is often obvious anyway. Nothing about the image is a secret, and a short note removes the awkwardness of someone asking. What matters more is that the faces are honest and that everyone in the photo agreed to appear.

Why does the AI put the wrong number of fingers on hands?

Hands are small, high-detail and rarely identical between images, so the model has less consistent information to draw from than it does for a face. It is the same reason jewellery and spectacle arms deform. Checking the hands at full size before you send, and regenerating if they are wrong, is the practical answer.

Can I check whether a photo someone sent me was AI?

Look for the tells rather than trusting a feeling. Start with the hands, then read the background text, then check whether the light in the picture has a source. Remember that the face being convincing proves nothing — a real face can be placed into a generated scene, so check the scene rather than the person.

Is it dishonest to use AI for a Christmas card?

Not on its own. It becomes a problem when the image claims something untrue about where people were or who was present, or when it uses someone's face without their agreement. A clearly festive portrait built from a real photo, sent with a note, is not a deception — it is a card.

What are the tells people wrongly treat as proof?

Blur, grain, uneven lighting, a soft low-resolution file and a tidy room are all things real photographs have, and none of them points to a generated image. The reliable tells are structural — hands, teeth, jewellery and background text — rather than atmospheric.

Does the same check work on a video?

Partly. Check the hands and any text first, then watch how a face moves between frames: features that shift slightly, a mouth that does not match the words, or a background detail that changes when the camera moves are the video versions of the photo tells. A single still frame from a generated clip can pass every test a still photograph passes.

Can an AI detector tell me whether a photo was generated?

Treat detectors as unreliable in both directions: they miss generated images and flag real photographs, particularly ones that have been heavily edited or compressed. The structural tells are what you can check yourself, and on a single image a person looking closely at hands and background text is often the more accurate judge.

→ Make a Christmas photo from a real photo · Generate a festive scene · Create a free account

Is This Christmas Photo AI? How to Tell, and How to Make Honest Ones