How Our AI Predicts Baby Gender From Your Ultrasound

AI predicts baby gender the same way an expert sonographer does — by reading the genital tubercle on a side-on (mid-sagittal) 12-week scan — but it does it consistently and at scale. The classic measurement comes from Efrat et al. (1999): a nub angled more than 30° from the lower body suggests a boy, while a nub lying parallel or under 10° suggests a girl. Angles between 10° and 30° are indeterminate.

Instead of eyeballing that angle, NubHub’s model evaluates several signals at once — the precise tubercle angle, the shape of the nub, the “stacking” sign, and how clear the image actually is — and combines them into a single confidence score. If you want the underlying method explained from scratch, start with our complete guide to nub theory.

Nothing is wrong with a skilled reader — the problem is that human reading alone is subjective and inconsistent. Show the same blurry 12-week scan to two experienced readers and they can genuinely disagree, because the nub is only a few pixels wide and the angle is a judgement call.

• Subjectivity: where one reader sees a 28° “girl-ish” nub, another sees 32° and calls a boy.
• Fatigue: a person reading their hundredth scan of the day is not as sharp as on their first.
• Inconsistency: the same reader can even call the same image differently on a different day.

That variability is exactly why honest accuracy for nub reading sits around 85–90% at 12–13 weeks with a clear image (Manzanares et al., 2016) rather than the “99%” some sites claim. Untrained guessing at home is far less reliable still. Our accuracy guide breaks the numbers down.

The process is deliberately simple for you and rigorous behind the scenes:

1. Upload your 12-week scan. A clear mid-sagittal (side-profile) image works best — see our guide to reading a 12-week scan for what that looks like.
2. The AI analyses it. The model assesses the nub’s angle, shape, stacking and overall image clarity together, rather than one factor in isolation.
3. It produces a confidence score. A crisp, well-angled nub scores high; a blurry or awkwardly positioned one scores lower and is flagged.
4. A real person double-checks it. A human reviewer looks over the AI’s reading, catches edge cases, and confirms the call.
5. You get your result in ~2 hours. Backed by our money-back guarantee.

Because the model gives a confidence score rather than a flat yes/no, you also learn how sure the reading is — which a single human glance rarely tells you.

The common objection is “human eyes beat AI.” Treated fairly, the honest answer is: it is not AI versus humans — it is AI and a human double-check, which is more reliable than either alone.

Each side covers the other’s weakness. AI is consistent and never fatigues — it applies the same measurement to every scan, all day, without drift. A human reviewer brings judgement on edge cases the model hasn’t seen cleanly — an unusual fetal position, a cord mistaken for a nub, a borderline angle. Pairing a tireless, consistent measurer with a second pair of human eyes is simply more dependable than relying on either by itself.

That is why every NubHub reading pairs the AI with a human double-check before it reaches you.

AI for fetal-sex classification is a genuine and emerging research area, not yet a regulated clinical tool. The strongest published result so far comes from a 2024 study in Expert Systems with Applications: a DenseNet201 deep-learning model reached 96.27% accuracy classifying fetal sex from ultrasound images, rising to 97.82% when combined with a linear SVM, on a dataset of 4,400 expert-labelled images (Expert Systems with Applications, 2024).

Two honest caveats matter here. First, these are research datasets, not regulated medical devices — promising signals, not proven medicine. Second, real-world scans vary far more than a curated dataset, which is precisely why NubHub keeps a real person in the loop. We frame AI as a powerful, improving tool — never as a diagnostic one.

For a genuinely confident reading, yes. A DIY guess has no method and no second opinion; most free apps are thin novelty tools with no human check and no accountability. NubHub adds the two things that actually move accuracy: a consistent AI measurement and a human double-check.

If you simply want to see boy-nub vs girl-nub features for yourself, our boy vs girl nub guide walks through labelled examples.

Your scan is handled through a private flow and used only to produce your prediction. We do not share or publish your scan without your permission, and it is never sold or posted to social channels on your behalf.

In short: your image goes in, the AI reads it and a real person double-checks it, your result comes back to you — and that is the full extent of how the scan is used.

No. NubHub’s AI is an entertainment service, not a medical or diagnostic tool. It offers an early, for-fun estimate of your baby’s sex from the nub — it does not diagnose anything and is not a substitute for clinical care.

The medically validated methods remain NIPT (a blood test from around 10 weeks) and the 20-week anatomy scan. Treat your NubHub result as an exciting early glimpse to enjoy while you wait for those. Questions? See our FAQ.