How Accurate Is a Baby Face Preview
It can look convincing and still be wrong. Accuracy here means “does this resemble the photos,” not “will the child look like this.”
What the model actually uses
Both photos are sent as references with a written prompt. The prompt asks for a photo-realistic boy about 1 year old, then a girl of the same age, with skin tone and features blended evenly from both adults and with different clothes and background colours. A style note is appended as extra direction.
Eye color and hair color in a real child depend on many genes and on chance. A photo model does not measure those genes. A blue-eyed illustration is a drawing choice.
What to tell guests
Say it is for entertainment. Do not tell someone the picture is a genetic prediction, a paternity opinion, or a medical sex result.
If a portrait looks nothing like the parents, that is a miss, not a billable defect. Run it again with clearer photos, or stop. Credits come back only when the provider fails.
Safety filter
We do not run face recognition to guess ages. The generation provider can still refuse a photo. A refusal fails the run and returns the credit.
Do not upload minors. The checkbox is there because the adults in the photos, and you, need to have agreed.
FAQ
- Does the preview read DNA?
- No. The model sees the two photos you upload. It does not see a lab sample, a family tree, or medical records.
- Why do two runs look different?
- Each run is a new illustration. Lighting, crops, and the style note change the picture. That variation is expected.