Describe the animation of the photo. Return JSON {"photo":"...","motion":"...","cameraMove":"...","prompt":"..."}: photo - what’s in the photo (briefly), motion - which MICRO-movement to animate and what remains motionless, cameraMove - camera movement (can be “static”), prompt - final instruction text for revitalization.
Reviving a photo works differently than generating it from scratch: the neural network takes your picture as the first frame and completes the next ones, trying not to destroy what is already there. Therefore, here you are not describing the appearance - it is already given - but you are describing ONLY the movement: what should move and how the camera behaves. What comes to life easily: hair and fabric in the wind, water, smoke, steam, clouds, a slight turn of the head, blinking, slow movement. What breaks: complex hand gestures, a sharp turn of the whole body, the appearance of new objects, text and small faces in the background. Insider: ask for a micro-movement, not a plot. “A light breeze moves the hair, the camera barely zooms in” will give a lively portrait; “the person gets up and leaves” will turn the face into mush, because the model is conjecturing something that is not in the photo. The second technique: clearly indicate what remains motionless (“the tree trunk is motionless, only the leaves move in the wind”) - this is the anchor that holds the picture. A common mistake is to load a photo with a long action scenario. The more modest the movement, the more believable the animated frame.
Unlock access to submit solutions for instant AI review.