PredictMyBaby generates images of a potential child from a front-facing photograph of each parent, using a GAN that reads more than 70 facial landmarks from both and blends the traits. Output comes in sets showing the child at different ages and either gender, and the company reports more than 58,800 generations for 2,540 customers.
The accuracy claim needs addressing directly, because the page makes two incompatible statements. It advertises 98 percent accurate baby predictions, and further down concedes the result is not an exact prediction since real genetics are complex and include hidden traits. Only the second is true. There is no way to measure the accuracy of a picture of a child who does not exist, so the 98 percent is not a figure that could be right or wrong – it is decoration on a novelty product, and the honest framing is entertainment.
Taken as entertainment it is straightforward and fairly priced: three free generations with no account, then Basic at $9.99 for eight photos, Premium at $19.99 for 24 across both genders, and Elite at $39.99 for 48 across multiple ages, all one-time payments with no subscription. It requires uploading identifiable photos of two real people, a generated face resembles the average of its inputs rather than any actual genetic outcome, and expectant parents are a receptive audience for a confident-sounding number.
