When Faces Change in an AI Couple Video
A face can look different during movement, or two subjects can appear to merge during an embrace. Start by locating the change in the finished clip. The checks below are suggestions for reviewing inputs and results, not a proven fix or a promise of a free remake.
Published: · Updated:
Check the result before retrying
Confirm which photo belongs in each role, whether the intended face is clear in each crop, and where the change begins. If the task is still processing or its state is unclear, check the existing task before submitting another one.
- A visual mismatch and a failed generation are different outcomes.
- Higher resolution, a different model or another attempt does not guarantee a matching face.
- Record the task ID and problem timestamp when asking for help.
First identify what went wrong
Photo rejected before submission: follow the file-format, size or decoding message beside that photo. This is an upload or input problem, not face drift in a generated video.
Task still processing or status uncertain: open the existing task and check its state. A wait or missing response does not establish that generation failed. Avoid a duplicate submission while you are checking.
Task completed, but a face changes: review the downloaded clip and note the first affected frame or timestamp. That is a result-quality issue, even when the generation completed successfully.
Review photos and story settingsCheck the photos and their roles
For a couple story, the first photo is the Survivor and the second is the Zombie Partner. For the pet story, the first is the adult owner and the second is the dog. Confirm the visible labels after changing stories or exchanging photos.
Review each crop: is there one intended subject, with a visible face and enough detail to recognize them? Try replacing a dark, distant, heavily filtered or obstructed photo with a clearer authorized one. This is an input check, not a guarantee that the next result will match.
Find the point where identity changes
At the first appearance: compare the generated subject with the intended photo and confirm the roles. A mismatch from the start calls for a different review from a face that changes only during contact.
During the hug: check whether the faces become hidden, overlap or appear fused. Note the timestamp and which person is affected. For a dog story, inspect the dog’s muzzle, markings and paws as well as the owner’s face.
At the memory cut: compare the face, hair, clothing and subject position before and after the cut. Record whether only the transition changes or the subject remains different for the rest of the video.
Choose one change to evaluate
On this website, you choose prepared templates and model settings; you do not directly edit the underlying story prompt. Start with the photo or role issue you identified. Keep a note of the original settings so you can compare results if you decide to spend credits on another attempt.
In a manual tool that accepts prompts, you can experiment with distinct subject roles, a simpler action and an angle that keeps both faces visible. These are suggestions to test. They are not a verified before-and-after repair method for this website.
Review the cost and failure policy before a retry
Another generation can consume credits. Review the new quote with your balance before confirming. Changing model, duration or resolution can change the cost without ensuring a better likeness.
Failed charged tasks return eligible credits under the published rules. A technically completed result with an unwanted face or motion detail does not automatically qualify for a free remake or cash refund. Check the task and credit history before buying another pack.
Send enough detail for support to investigate
Include the task ID, selected story, model, duration, resolution and the timestamp where the problem appears. Describe whether the task failed, is still processing, or completed with a visual mismatch. If credits seem wrong, include the visible credit-history entry.
Use the support channel on the contact page. Do not post private input photos or private video links in a public forum. Sharing a task ID helps identify the record; it does not make the video publicly accessible or guarantee a refund.
Frequently asked questions
Will 2K fix a changing face?
There is no such guarantee. Resolution is an output setting. Review the actual result and current cost before another attempt.
Should I submit again if the task is taking longer?
Check the existing task first. Its status may still be processing or uncertain; submitting again can create another paid task.
Is every distorted result refunded?
No automatic refund is promised for an unwanted detail in a completed result. Failed charged tasks and cash refund requests are handled under the published policy.