Generation reconstructs a picture
An AI family portrait is a newly generated image. It is not a pixel-for-pixel cutout of each uploaded face placed on a different background. Changing the pose, light, camera angle or artistic style can change visible facial details. Even a polished result can drift away from the person in the source. Treat likeness as something to inspect carefully, not something that follows automatically from a prompt asking the model to preserve identity.
A reference can hide the information you need
A distant face, heavy filter or obstructed eye leaves less visible information to work from. When you look at a result, ask which distinctive details were actually clear in the original. A face seen only in profile cannot fully describe the other side. Upscaling a tiny source may make its edges look smoother without recovering the true shape of the eyes or mouth. Start with a clearer original rather than adding increasingly forceful identity instructions.
Multiple people increase the review work
For group portraits, keep each subject’s references together in a separate upload slot. Two photos in one slot describe the same person. Mixing family members inside a slot can make the intended assignment unclear. Confirm the expected people and pet counts before submission, then count everyone in the output. Correct headcount and recognizable identities are different checks: a result can contain the right number of people but still change one person’s features.
Style and pose change what a face looks like
An illustrated portrait intentionally changes texture, edges and proportions. A 3D cartoon and a photographic studio portrait therefore need different expectations. A dramatic expression or large camera-angle change also changes which features are visible. If resemblance matters more than a creative transformation, begin with a straightforward photographic scene and a natural expression. Keep the originals beside the result so you can judge identity separately from the appeal of the chosen style.
Compare details instead of overall polish
Review one face at a time at a useful zoom level. Look at face width, eyebrow shape, eye spacing, nose shape, smile, hairline and any distinctive marks that are visible in the original. Check age, build and skin tone too. Avoid judging only by the first impression of a well-lit photograph. Ask another family member to review the portrait privately if you are uncertain, and do not present an uncertain reconstruction as a documentary photograph.
Change the source before another attempt
If the result changes a person substantially, replace that person’s reference with a clearer photo or a more compatible angle. Repeating exactly the same setup does not guarantee an improvement. Keep a note of the scene, references and output you compared so you can tell whether a new attempt is useful. The replacement-photo redo for eligible paid jobs is limited and does not promise a closer likeness. Review its current eligibility on the result page.
Separate technical failure from dissatisfaction
A provider error and a completed photograph you dislike are different outcomes. Technical failures follow the system retry and credit-return path. A completed result normally uses the stated credits, even when you prefer the source face. Check your task and credit history before starting another job after a connection problem: the original may still be running. Recovering an existing job is safer than repeatedly submitting a new one.
Edits and enlargement have limits
Changing a background or outfit creates another generated version; it can affect areas you hoped to preserve. Enlargement increases output dimensions but cannot prove a face is more accurate. Review every edited or enlarged version again before sharing or printing. Keep the original result and the references so you can return to a version you prefer. For sensitive memorial work, label the result as an AI-created remembrance rather than evidence of an event.