About Cradlepeek
Cradlepeek is a small web tool that turns a photo of each parent into a portrait of a child who could be theirs. The Cradlepeek team builds it, tests it and answers the mail. We keep it simple, and we say plainly what it cannot do.
By The Cradlepeek team (Product and testing). Last updated .
Who runs Cradlepeek
The Cradlepeek team is a small group that builds single-purpose web tools. We write the prompt, pick the limits and read every message from the contact form. So when a page here says "we", it means the people who also fix the bugs.
We are not geneticists, and the tool is not a genetic test either. An image model makes each portrait from your photos and from a prompt that we wrote. That is why we call the result an imagined likeness, never a prediction.
How the tool works
You upload a photo of each parent, or just one. Then you pick an age, from newborn to 13 years, and a girl, a boy, twins or a surprise. Our server shrinks a copy of each photo and sends both to Google Gemini 2.5 Flash Image, together with a fixed prompt.
That prompt asks the model to mix face shape, eyes, nose, lips, hair and skin tone from both parents. It also asks for exactly one child, a real photo look and no text in the picture. You never write a prompt yourself.
How we test
We test with four photos: two illustrative couples, a mom and a dad each. An image model made all four from a text prompt, so they show no real person. As a result, nobody's face sits on this site without consent.
On October 19, 2026 we made twelve child portraits from those photos with Gemini 2.5 Flash Image. The site has no model key yet, so we ran them through a Freepik account. However, each prompt was exactly the text that the code on this site sends. The set covers every age, twins and the one parent mode. Nothing was edited or run twice.



Then we compare each portrait with the parent photos by hand. We check five things. First, the picture must show one child, or two for twins. Second, the child should resemble both parents. Third, the age must look right. Fourth, the picture should hold no text or watermark. Last, it must look like a real photo, not a painting.
What the tests showed
Not every result passes cleanly, and we say so. With our first couple, most children mix both parents well. With the second couple, however, the model leaned heavily toward the mom. Their newborn, toddler and 6 year old show her dark skin and coily hair, and the dad barely shows.
The twins of the second couple missed too. Instead of two blends, one twin copies the mom and the other copies the dad, so they look unrelated. In addition, the 13 year old boy looks closer to 15 or 16, and one 6 year old has no gap in her teeth although the prompt asks for one. Finally, we have not timed the tool yet, which is why no page quotes a speed.
Your photos
Our server keeps both photos in memory for one request and forgets them as soon as it sends the portrait back. The photos do travel to Google for that request. However, we keep no copy, build no gallery and train nothing on your pictures. You can read the details in our privacy policy.
What the tool is built with
The site runs on Next.js. The open-source library sharp reads your uploads, shrinks them and adds the watermark. Google Gemini 2.5 Flash Image makes the portrait. Because we rely on these projects, we link to them below.
Get in touch
Do you have a question, a bug report or a partnership idea? Then write to us on the contact page, because we read every message.