The prompt we give our video model
Our model gets your photo as the identity and the original clip as the motion, with this instruction: replace the man behind the lit window with the main subject of the photo; the face must be the face from the photo (the exact facial features, with the hair and outfit from the photo), not the face of the man in the clip; the expression, facial movements and actions stay exactly the same as in the clip, with the same timing: eating the burger slowly while staring into the camera, noticing the camera, pointing at it and pulling the curtain closed; and keep everything else the same: the night apartment building, the camera zooming in to the lit window, the paper bag and wrapped burger on the table.
A prompt for general AI video tools
Outside view looking into a lit apartment window at night. First the person stares directly at the camera with the curtains open. Then they raise one hand and point at the camera. Only after pointing do they lower the hand, grab the curtain and pull it completely shut. One continuous shot. Each action finishes before the next begins. End with the curtain fully closed.
A general tool does not have the original clip to copy, so it may skip the point, close the curtain early or melt the burger into the hand. Those are model limits, not a prompt mistake; generating shorter clips and joining them is the usual workaround.
Which to use
Upload one photo here if you want the original scene and timing without writing anything. Use a general tool and a prompt if you want something different from the trend, such as another food or another setting.
Last checked .