You typed "cute newborn baby photoshoot, professional photography" and got back a plastic doll. Or worse — a 35-year-old man's face shrunk down onto an infant body, propped in a basket, with seven fingers on one hand. Every AI image model does this. It's not you.

What an AI baby photoshoot prompt actually needs

An AI baby photoshoot prompt works when it specifies four things the model won't infer on its own: the baby's exact age (in days, weeks or months), the pose by its photography name, one soft directional light source, and the fabric by name. "Newborn baby, cute, professional" gives the model nothing to anchor on, so it defaults to adult facial geometry and studio-lit plastic. "14-day-old newborn, curled taco pose, cream stretch-knit wrap, single soft window light from the upper left, 85mm at f/2.0" gives it four constraints, and constraints are the whole game.

Why the faces come out wrong

Image models learn faces from the internet, and the internet is mostly adults. When you write "baby," the model does something closer to scale down an adult than render an infant. That's why AI babies have long noses, defined jawlines, and deep-set eyes — all adult features.

Infant proportion is genuinely different, and you have to say it out loud:

Feature What AI defaults to What you should write
Forehead Adult ratio, ~1/3 of face "large rounded forehead, cranium taking the upper half of the head"
Eyes Deep-set, defined brow "low-set eyes, minimal brow ridge, slightly puffy lids"
Nose Long, bridged "small button nose, no bridge"
Jaw Visible jawline "no visible jawline, soft rounded cheeks and chin"
Skin Smooth and matte "mottled newborn skin, fine downy hair, slight peeling on hands"

That last row matters more than people expect. Real newborns aren't airbrushed — they're blotchy, a little red, sometimes peeling. The word mottled does more for realism than any camera setting you can name.

Age is the highest-leverage token in the whole prompt. "Newborn" is vague. "5-day-old" is a specification. A 5-day-old, a 3-month-old and a 9-month-old are three completely different subjects, and models genuinely respond to the distinction.

Newborn photography is a lighting genre, not a subject

Look at any newborn photographer's portfolio and you'll notice the light is always the same: one big soft source, usually from above and slightly behind the baby's head, feathered so it falls off toward the feet. That's the entire look. It's why the images feel calm.

AI models default to flat, even, frontal light — which reads as "stock photo" instantly. So name the light:

  • One source, not three. "Single soft window light" beats "beautiful studio lighting."
  • Direction. "From the upper left, angled down toward the crown of the head."
  • Fall-off. "Light feathering off toward the feet, gentle shadow in the lower frame."

Add "soft neutral palette — cream, oatmeal, dusty sage" and you've locked the colour grading too. Newborn photography lives in beige for a reason: it keeps attention on skin.

The pose vocabulary models actually recognise

Newborn photographers have names for shots, and image models have seen those names captioned thousands of times. Using them is free accuracy.

Poses that render well: taco pose (curled, knees to chest), potato sack (wrapped upright, arms in), side-lying curled, on-the-back-with-hands-under-chin, wrapped in a bowl or basket, parent's-hands-cradling.

Poses to avoid asking for: the "froggy pose" (chin on hands, elbows out) is a composite in real photography — two images blended, with a spotter's hands edited out. Ask AI for it and you'll usually get a dislocated-looking arm. Skip it.

One more trick: hands are where AI fails, so pose them away. Wrapped, tucked under the chin, curled into fists, hidden in a blanket. You'll dodge the extra-finger problem entirely rather than fighting it.

The prompts

Swap anything in [brackets]. Each of these works in ChatGPT Image, Midjourney, Flux and Nano Banana — notes on model differences are in the table further down.

1. The base studio recipe

A 12-day-old newborn, curled taco pose, swaddled in a cream stretch-knit wrap,
resting on a textured oatmeal fabric backdrop. Large rounded forehead, small
button nose, no visible jawline, soft mottled newborn skin with fine downy hair.
Single large soft light from the upper left, feathering off toward the feet.
Shot on 85mm at f/2.0, shallow depth of field. Soft neutral palette — cream,
oatmeal, warm beige. Photorealistic, calm, quiet mood.

Why it's built this way: age → proportions → fabric → single light → lens. Every clause removes a decision the model would otherwise make badly. Swap: the age (try 3-week-old vs 6-week-old and watch the face change), the wrap colour, and the backdrop texture.

2. Lifestyle, in-home, natural light

A 5-week-old baby lying on unmade white linen bedding in a sunlit bedroom.
Documentary lifestyle newborn photography, not a studio setup. Soft morning
light through a sheer curtain, window out of frame to the right, gentle
shadows across the sheets. Baby in a plain white onesie, arms loose,
eyes half-closed. Muted warm tones, slight film grain. 35mm at f/1.8,
natural and unposed.

Why it's built this way: "documentary lifestyle, not a studio setup" is a negation that actively pushes the model out of its stock-photo default. "Unmade bedding" and "arms loose" buy you the imperfection that reads as real. Swap: bedroom for nursery, morning light for late-afternoon, white linen for a knitted throw.

3. Macro detail shot — feet, hands, eyelashes

Extreme close-up macro of a sleeping 9-day-old newborn's feet, cradled in an
adult's open palms. Wrinkled soles, tiny toes slightly curled, fine skin
texture and faint peeling. Warm skin tones against a dark blurred background.
Single soft side light raking across from the left. 100mm macro lens, f/2.8,
razor-thin focus on the toes. Photorealistic, intimate, minimal.

Why it's built this way: macro shots are the safest way to get anatomy right, because there's less of it in frame. "Raking side light" is what makes skin texture visible at all — flat light erases it. Swap: feet for a hand curled around a parent's finger, or a close-up of eyelashes against a cheek.

4. The milestone / sitter session (6–9 months)

A 7-month-old baby sitting upright unassisted, wearing a cream knit romper,
on a plain sage-green seamless backdrop. Chubby cheeks, gummy open-mouth
smile with two lower teeth, hands resting on knees. Large soft light source
front-left with a subtle reflector fill on the right. 85mm at f/2.5,
eye-level with the baby. Clean, bright, joyful, photorealistic.

Why it's built this way: "sitting upright unassisted" and "two lower teeth" are age markers — they stop the model from splitting the difference between newborn and toddler. "Eye-level with the baby" is the single most-forgotten camera instruction and it changes everything. Swap: the backdrop colour, the outfit, "gummy smile" for "serious curious expression."

5. Cake smash / first birthday

A 12-month-old sitting on the floor mid-cake-smash, small pastel buttercream
cake half-destroyed, frosting on both hands, one cheek and the chin, laughing
with eyes squeezed shut. Pastel balloon and streamer backdrop, soft even
frontal light, confetti mid-air. 50mm at f/2.0, slight motion blur in the
hands. Bright, candid, messy, joyful.

Why it's built this way: cake smash prompts fail because models render a tidy child beside an untouched cake. You have to specify the mess in three places — hands, cheek, chin — before it commits. "Slight motion blur in the hands" sells the candour. Swap: the palette, "laughing" for "startled," balloons for a floral arch.

6. Newborn with sibling

A 4-year-old girl sitting cross-legged on a bed, carefully holding her
2-week-old newborn sibling in her lap, both in matching cream knitwear.
Her chin tilted down, looking at the baby, small proud smile. Newborn
asleep, hands tucked under the chin. Soft window light from behind them
creating a gentle rim on their hair. 50mm at f/2.2, warm neutral tones,
tender and quiet.

Why it's built this way: two subjects means twice the chance of anatomy errors, so both sets of hands are constrained — hers are "holding," the baby's are "tucked." The gaze direction is stated, which is what makes the image feel like a relationship rather than two people in a frame. Swap: the sibling's age, "sitting cross-legged" for standing and cradling, the light for a warm lamp.

7. Themed / whimsical setup

A sleeping 3-week-old newborn nestled inside a woven basket lined with
faux fur, surrounded by dried eucalyptus and cream roses. Wearing a soft
knitted bonnet. Overhead-angle shot looking straight down. Diffused
overhead light, gentle vignette in the corners. Muted earthy palette —
moss, bone, dusty rose. 50mm at f/2.8, dreamy and painterly but
photorealistic.

Why it's built this way: themed setups drift into cartoon fast, so "photorealistic" has to be the last word after "painterly." The overhead angle is stated explicitly because models default to eye-level. Swap: the props (a crescent-moon prop, a knitted nest, a vintage suitcase), the florals, the palette.

8. The fix-it prompt

[Paste your working prompt]

Negative: extra fingers, fused fingers, malformed hands, adult facial
proportions, defined jawline, doll-like plastic skin, airbrushed skin,
teeth on a newborn, harsh direct flash, multiple light sources, oversaturated
colours, text, watermark.

Why it's built this way: this is the block you append when the base prompt is 80% there. "Adult facial proportions" and "defined jawline" do the heaviest lifting — most people only negate hands and wonder why the face still looks wrong. Swap: nothing. Keep it as a snippet and paste it every time. (Note: Midjourney uses --no rather than a Negative block, and ChatGPT Image ignores negatives — just say it as an instruction instead.)

Which model for which shot

Model Infant proportions Hands & feet Light quality Best for
Flux Very good — holds real anatomy Best of the four Naturalistic, slightly flat Macro details, lifestyle, anything with visible hands
Nano Banana Very good Good Clean and modern Studio setups, edits to an existing image, quick iteration
Midjourney Tends to age the face up — needs the proportion clauses Fair Gorgeous, the prettiest of the four Themed and whimsical setups, styled florals, editorial looks
ChatGPT Image Fair, drifts toward doll-like Fair Soft, sometimes waxy Conversational refining — "make the light softer, lower the camera"

If you only run one: Flux for realism, Midjourney for beauty. For anything close-cropped on hands or feet, go Flux and don't argue with it.

One honest caveat

These prompts generate images of babies who don't exist. That's the point of them — mood-boarding a real session, testing a theme before you book a photographer, making an invitation or announcement backdrop.

What they can't do is photograph your baby. And it's worth being deliberate about uploading real photos of a child to image tools you haven't read the terms of — some retain and train on uploads. If the goal is a keepsake of an actual child, the AI's real job here is the shot list you hand to the person with the camera.

Related guides

Steal the rest

Every prompt above is a starting point, not a finished thing — the good ones come from swapping one variable at a time until it clicks. The full set of baby and newborn prompts, tested and tagged by age and setup, lives here:

→ Baby Photoshoot prompts on PromptThisOne

Or browse the wider Portraits collection if you're shooting the whole family.