The Best Photos for AI Dance Videos (And the Ones to Avoid)

AI motion transfer copies identity from your reference photo — face, hair, build, outfit. Whatever is in (or missing from) the photo ends up in the video. A minute spent choosing the right picture saves a disappointing render.

What works

• Even, frontal light. Shadows across the face become shadows in every frame.
• One subject, facing the camera. The model needs an unambiguous face to lock onto.
• Head to waist or knees. The dance is a full-body performance; more outfit in means more outfit out.
• Plain, high-resolution files. JPG, PNG, or WebP up to 10 MB, at least 300×300 pixels.

What to avoid

• Mirror selfies. The phone is a face-shaped object the model will happily render.
• Group photos. Two faces means a blended stranger, not two characters.
• Sunglasses, hats, hands over the face. Occlusion degrades identity transfer.
• Screenshots of screenshots. Compression artifacts amplify across every generated frame.

The two-photo rule

For the couple version, use one photo per performer: the little man in slot one, the princess in slot two. Separate references keep each character's face and build distinct — which is the entire joke of the second character's reaction.

Pick well, and the barn does the rest. Start at the generator.