Every December someone in the family group chat asks for "one of those AI Christmas photos". And every December the first attempt comes back with Mum looking like a stranger and an extra kid nobody recognises.
The fix is boring but it works: start with the right photo, use a prompt that's specific about the scene, and tell the app plainly not to touch the faces. Here's how I'd do it in Gemini or ChatGPT, with prompts copied from real pages in the Promptclay Christmas photo prompts collection.
1. The photo matters more than the prompt
The app copies what it can see. So before writing anything, look at your photo:
- Is everyone in it? It can't add someone who isn't there.
- Are the faces big enough? A wide shot from across the room leaves it guessing.
- Any sunglasses, low hats, hands over faces, people turned away? Each one is a face the app has to make up.
- Is it sharp? A blurry person tends to come back as a lookalike.
A quick, close phone photo beats a pretty one taken from far away. Smaller groups also hold up better. If you have a big family, make two images and put them together on the card.
2. Match the scene to your pose
Faces slip most when the app has to move bodies around a lot. If everyone's standing in your photo, a standing scene will keep faces closer than one where you all end up cross-legged on a rug.
Each prompt below has a blank in curly brackets. Fill it with who's actually in your photo, like "a couple and their two grown sons".
Matching pyjamas by the tree

Example image from the Family in Matching Christmas Pajamas by the Tree prompt page (GPT Image 2.5 in ChatGPT).
Real photograph of {members} in matching red plaid Christmas pajamas, sitting on a rug in front of a lit tree on Christmas morning, mugs in hand, wrapping paper around them. Soft window light, warm smiles, slightly playful poses. Eye-level, 4:5 portrait framing. "Real photograph" plus "soft window light" keeps it looking like a phone photo instead of a glossy render.
Card portrait by the fireplace

Example image from the Family Christmas Card Portrait by the Fireplace prompt page (GPT Image 2.5).
Studio-quality family Christmas card portrait of {members} posed in front of a decorated mantel and fireplace, coordinated in deep green and cream knitwear, relaxed smiles, garlands and warm string lights behind them. Soft key light, shallow depth of field, clean space at the top for a greeting. 4:5 portrait. Two details carry this one. Naming two colours gets you outfits that actually match. And "clean space at the top for a greeting" means your "Merry Christmas" won't land on someone's forehead.
Trimming the tree
Candid real photograph of {members} trimming the Christmas tree together in a cosy living room: one hangs a glass ornament, one lifts a string of lights, one holds the star, everyone laughing in knit sweaters. Warm lamp light, lit tree, boxes of ornaments on the sofa, a fire glowing behind. Eye-level, wide enough to show everyone, documentary style. From the Family Trimming the Christmas Tree Together page. Giving each person a task stops the app lining you all up like a passport queue. Good for casual photos where people aren't all facing the camera.
3. The line that keeps the faces
Add this to the end of any prompt above:
Use the people in my uploaded photo. Keep every face, age, skin tone, hairstyle and the number of people exactly the same. Do not add or remove anyone. It covers the three things these apps get wrong most: who to use, what to leave alone, and the head count. It won't make results perfect. You still need to check each face.
4. Running it
Open a fresh chat so old instructions don't leak in. Attach the photo, paste the prompt with the blank filled in, add the face line, send.
Which app? Both take a photo plus a prompt. In what we've seen, Gemini tends to hold faces closer on group edits, and ChatGPT tends to follow outfit and layout details more tightly. If faces are the priority, start with Gemini. If the card layout is, start with ChatGPT. Running the same prompt in both costs you a minute.
If there are children in the photo, use a parent's account and check the app's rules first.
5. Fix one thing per message
When the result is nearly right, don't rewrite the prompt. Reply with a single fix:
- "Make Dad's face match my photo exactly. Change nothing else."
- "There should be four people, not five."
- "Same image, warmer light."
Ask for three things at once and the app often regenerates everything, including the faces it had already nailed.
Want your exact photo, just festive?
If you don't want anyone re-posed, use a transform prompt that keeps your layout and camera angle. The Image Transformation to Cozy Christmas Scene prompt does that, adding lights, a tree and gifts around what's already in your photo. Faces usually survive better this way. Your living room still looks like your living room, though.
For more scenes, including pets in Santa hats, couples and snowy markets, the Christmas photo prompts collection on Promptclay has them all with the example image next to each prompt. They're free to copy and there's no sign-up.

