Separate identity change from a new expression
A person can look different when the camera angle, lighting or expression changes. Face drift is a stronger inconsistency: facial proportions, hairline or identifying details change across frames. A single attractive still does not prove that the whole video stays consistent.
Compare the beginning, middle and end with both original photos. Check left/right identity, eyes, jawline, hair and clothing. Hands moving over a face or another subject crossing it can make comparison harder.
Improve the next attempt deliberately
Choose a clearer original with one adult, visible facial detail and even light. Use similarly framed images for the pair and avoid strong filters or heavy occlusion. Confirm that you did not accidentally put the same reference in both slots.
Try a short clip with a relaxed vibe before a longer performance. Change one useful input at a time, then compare results. A different quality tier changes requested resolution and cost, not a guaranteed identity score. Re-running repeatedly is not a reliable fix.
Know which tasks need support
A completed clip with imperfect resemblance still consumed a generation. Confirmed failed tasks return reserved credits; a completed aesthetic result and a failed task are different states. Check the task in your private library before starting another attempt.
If a request has an uncertain submission state, contact support with the task reference. Do not upload someone else's face without permission or describe generated output as proof of a real event.