Pick a Full Photo Instead of a Small CropA small crop would not have enough visual information for the AI to work with. A normal photo usually gives better results.
Be Specific With the Motion You WantYou can describe exactly what your subject should do. For example, person on the left must turn and smile. This is better than a simple 'smile'.
Watch Out for ReflectionsSometimes you may not have the photos scanned on your phone and would probably have them framed on the wall. When you use photos of those framed photographs, make sure to avoid reflections as that would confuse the AI.
Do Not Cram Too Many Actions Into One PromptDo not ask for too many things at once like smiling, waving, turning, and zooming. That's because the video is only 5 seconds, and there's only a couple of actions it can do naturally during that time.
Avoid Nudity and NSFW PhotosNSFW photos may be blocked or fail moderation checks, so use safe images if you want the generation to go through reliably.