Troubleshooting
How to approach face drift in an AI duo video
Find where the likeness changes and make a focused adjustment before another attempt.
Written and maintained by Alors On Danse AI Video · Updated
Locate the change first
Face drift describes a likeness that changes during the clip. It can become more noticeable in profile, during turns, fast movement or occlusion. Watch the full output and note the time and affected subject. A single thumbnail can conceal the issue or exaggerate a brief change that needs to be assessed in motion.
Inspect the reference information
If the face is tiny, hidden by hair or heavily smoothed, try a clearer unobstructed photo. Verify the upload role and keep one subject in each reference. Adding aggressive sharpening or filters can introduce artificial texture. When a better original is available, use it instead of repeatedly processing the problematic image.
Change one main variable
Replace the affected subject’s photo while keeping the other image and aspect ratio as a comparison. Switching to 1080p is not a guaranteed likeness repair. A new generation may improve one part and introduce differences elsewhere. Each new task uses the credits required by its selected resolution.
Separate quality concerns from failure
A completed clip with an imperfect likeness is not automatically a failed task. Credits are returned for confirmed failed generations; a temporary network error alone does not establish failure. When contacting support, include the task ID, the time of the issue and a brief description. Do not send passwords or payment credentials.