Start by checking that each photo is clear and assigned to the intended side. Then confirm the reference clip matches the requested output length and inspect the full generated file.
If faces or sides are mixed up
Upload one person per photo.
Keep photo one for screen-left and photo two for screen-right.
Use sharp photos with faces visible.
If the scene or timing drifts
The reference video sets the requested scene, but an AI model may still change details. Compare the generated clip with the reference from beginning to end.
If the duration or audio is off
Check the actual file metadata and streams. A visible audio stream does not prove the original sound was preserved; listen to the result.