AI Baby Prompts for Gemini and ChatGPT: What We Tested
Illustrations are AI-generated
The AI baby prompt that works names both photos, says which parent is which, asks for a baby at a stated age and tells the model to blend the two faces. We ran eight generations in Gemini on 20 October 2026. Every prompt with an age gave a believable baby or toddler. The one phrased as a question, "What will our baby look like?", gave us an adult man.
Below are the exact prompts we used, what each one produced, and the settings that matter before you upload anyone's photo.
What is the best AI baby prompt for Gemini or ChatGPT?
The best AI baby prompt is an instruction, not a question. Tell the chatbot which uploaded photo is the mother and which is the father, give the baby an age, ask it to blend rather than copy, and fix the framing. In our test this prompt gave a consistent baby face in two out of two runs.
Upload one photo of each parent, then paste this:
Copy this prompt: "Image 1 is the mother, image 2 is the father. Create one photorealistic portrait of their baby at about 6 months old. Blend features from both parents instead of copying either face. Head and shoulders, looking at the camera, hands out of frame, plain light background, soft window daylight, sharp focus, natural skin texture, no text."
Each line does one job:
- "Image 1 is the mother, image 2 is the father" stops the model guessing who is who.
- "About 6 months old" sets the age. Without an age, the result can drift to any age, adults included.
- "Blend features instead of copying either face" pushes against a common failure, a baby that is one parent with a smaller nose.
- "Hands out of frame" avoids the extra or missing fingers image models still produce.
- "Plain light background, soft window daylight" keeps the picture about the face.
The same wording works in ChatGPT's image tool, because both chatbots read plain English. We ran our test in Gemini only, so the results below are Gemini results.
What happened when we tested AI baby prompts in Gemini?
We tested four AI baby prompts in Gemini on 20 October 2026, eight images in total. All four prompts with a stated age returned a plausible baby or toddler. The vague question returned an adult. Asking for one trait changed others. The parents were two fictional portraits we generated first, so no real person's face was used.
The model was gemini-2.5-flash-image, the model Google brands "Nano Banana", called through the API. The mother has light olive skin, green eyes and dark wavy hair. The father has medium brown skin, dark brown eyes and short black curls.
| Prompt | Runs | What came back |
|---|---|---|
| "Create a realistic image of the baby these two people would have together." | 2 | A smiling baby, dark curls, dark brown eyes. Both runs showed nearly the same face. |
| "What will our baby look like?" | 1 | An adult man with the mother's green eyes and the father's features. No baby. |
| The detailed prompt above | 2 | A baby with loose dark curls and brown eyes. Same face both times, different clothes. |
| The detailed prompt, aged 3 years | 1 | A toddler with the same curls and a longer face. |
| The detailed prompt plus "give the baby the mother's green eyes" | 1 | Green-hazel eyes, but straighter, lighter hair and lighter skin than the other runs. |
The surprise was how little the long prompt added over the short one. Gemini already produced a clean baby portrait from one sentence, as long as that sentence said "baby". The detailed version earned its keep on framing and on hands, not on the face.
None of the six runs without an eye-colour instruction gave the baby the mother's green eyes. Six images are far too few to call that a pattern. If a specific trait matters to you, the next section shows what happens when you ask for it.
Why did "What will our baby look like?" give us an adult?
"What will our baby look like?" gave us an adult because a question leaves the model to decide what kind of picture answers it. Gemini merged the two parents into one grown-up face and stopped there. An instruction with the word "baby" and an age leaves no room for that reading.
Questions are how people talk to a chatbot, so this mistake is easy to make. Rewrite the question as an order before you send it: "Create a portrait of their baby at about 6 months old." Every prompt in our test that was phrased as an instruction returned a baby or a toddler.
The same rule applies to follow-up edits. "Make the baby older" is weaker than "Show the same child at 3 years old, same face, same background."
Can you choose the baby's eye colour or hair in an AI baby prompt?
You can steer one trait in an AI baby prompt, but the model changes others along with it. When we added "give the baby the mother's green eyes", Gemini delivered green-hazel eyes. It also straightened the hair and lightened the skin, so the baby moved toward the mother's whole face rather than just her eyes.
A trait you type into the prompt is your choice, not a prediction. Real eye colour depends on the genes each parent carries, and a brown-eyed parent can still pass on green or blue.
To see the actual odds for a couple, the baby eye color calculator works them out from both parents' eye colours and is free to use. Prompt for the result it gives, and you have a picture that matches the most likely outcome rather than a wish.
How accurate is a ChatGPT or Gemini baby prediction?
A ChatGPT or Gemini baby is not a genetic prediction. The model has never seen your DNA. It reads two faces and draws a child who looks plausibly related to both. In our test the same prompt gave nearly the same face twice, which shows consistency, not accuracy.
Treat the picture as a likeness for fun. Siblings from the same two parents can look very different, and no image model can tell you which of those children you will have.
Dedicated baby generators work from the same idea with a fixed pipeline instead of a free-form chat. We compared several of them, ChatGPT included, in our tests of AI baby generators, with accuracy and pricing side by side.
Which AI baby prompts work for a newborn photoshoot look?
Newborn photoshoot prompts work best as a second step: generate the baby first, then restyle that picture. Tell the model to keep the face exactly as it is and describe only the scene. In our test Gemini turned the 6-month baby into a sleeping newborn in a swaddle and kept the face recognisable.
Copy this prompt: "Keep this baby's face exactly as it is. Restyle the photo as a newborn studio portrait: the baby asleep on a cream knitted blanket, wrapped in a soft sage-green swaddle, top-down view, soft window light, muted pastel tones, face in sharp focus, hands hidden in the swaddle."
Three scene swaps that use the same opening line:
- Moon and clouds: "…the baby asleep on a crescent moon pillow among soft white clouds, pastel night sky."
- Monthly milestone: "…the baby lying on a cream rug next to a wooden letter board that reads 6 MONTHS, top-down view."
- Flower crown: "…the baby asleep with a small crown of dried pastel flowers, cream background, close-up."
Keep "hands hidden" in every version, and check the spelling on any letter board before you share it. Hands and text are where image models still slip, and a newborn picture with six fingers ends up in nobody's family chat.
What photos should you upload with an AI baby prompt?
Upload one clear, front-facing photo of each parent, taken in even light, with no filters, sunglasses or heavy makeup. The model can only blend what it can see. A side profile or a dark selfie forces it to guess, and the guess is usually a copy of the clearer face.
A few checks before you upload:
- One face per photo. Group shots make the model choose, and it may choose the wrong person.
- Similar angle and distance for both parents, so neither face dominates.
- No beauty filter. Smoothed skin and enlarged eyes carry straight into the baby.
If a result looks off, the usual causes and fixes are in why your AI baby photo looks off.
Check the privacy setting too. ChatGPT and Gemini both have a setting that controls whether your chats help train their models. Turn it off before you upload faces, especially a partner's face.
When is a dedicated baby generator easier than writing a prompt?
A dedicated baby generator is easier when you want the result without writing or fixing prompts. On AI Baby Generator you upload one photo of each parent, choose the baby's gender and age, and get a photo in seconds. AI Baby Generator deletes uploaded photos within 30 days.
| Chatbot prompt (Gemini, ChatGPT) | AI Baby Generator | |
|---|---|---|
| What you do | Write and adjust a prompt | Upload two photos, pick gender and age |
| Age control | Only if the prompt states it | A setting |
| Typical failure | Adult result, copied parent | Weaker likeness from poor photos |
| Uploaded photos | Follow the chatbot's history settings | Deleted within 30 days |
| Cost | Your chatbot plan | Paid, starting with a 3-day trial |
Chatbot prompts are the better choice if you enjoy tinkering and already pay for a chatbot. For a quick, repeatable result from two photos, try the AI Baby Generator instead.
