Choosing photos for a two-character AI dance
A focused checklist for faces, framing, lighting and two distinct roles, with separate notes on cat and pet adaptations.
Updated
Prepare your two photos →Two-image dance generation has two jobs: keep each character recognizable and carry both through movement. Clear source photos reduce ambiguity, but do not guarantee a result. Choose photos for the entire sequence, including the performer who is smaller in the background.
One identity in each file
Choose one person per photo, with the face visible and large enough to inspect. Avoid collages, group shots, masks, sunglasses covering the eyes, and pictures where a hand obscures the face. If a group photo is your only source, obtain permission and prepare a separate crop of each intended adult before using it. Check A and B side by side. Their previews should make the cast unambiguous before you think about the dance. The input order defines the initial role, so swapping files is different from flipping an image horizontally.
Keep detail and natural light
Start with an original photo rather than a compressed screenshot or a heavily filtered social-media copy. Even lighting makes facial features easier to inspect. Avoid extreme backlight, motion blur, artificial face reshaping and strong colour effects. A JPG, PNG or WebP smaller than 10 MB fits the current local chooser. File size is a technical limit, not a quality score. Enlarging a blurry picture does not restore missing detail, and an unsupported file should be converted locally rather than merely renamed.
Framing matters more than matching the set
A natural front view or gentle angle is easier to inspect than an extreme profile. Include useful head and upper-body detail without making the face tiny. You do not need to photograph the person in front of a red wall: character identity and the dance setting play different roles. Do not expect the source outfit or background to be preserved exactly. Our planned workflow uses a fixed dance reference and does not expose a prompt editor or a catalogue of models.
What changes for cats and other pets?
A pet’s ears, muzzle, body shape and markings carry identity. A useful reference would show those features clearly with one animal per image. Two similar cats require careful checks in both close-ups and the shared dance; a model can merge or exchange their markings. Upright dancing is a stylized AI adaptation, not a recording of an animal performing. Our pet mode is not available, and success with human faces would not establish pet support. This guidance helps explain the format; it is not an invitation to submit pets to the restricted human workflow.
A final review before you publish
Check each role at the opening, after the exchange, in the wide shot and at the ending. Look for identity changes, implausible hands, merged bodies or unintended people. If the output is misleading or unsuitable, do not publish it. Keep permission from the people depicted and any rights needed for the photographs and music. A technically successful generation is not automatic permission to use someone’s likeness commercially. Return to the preparation guide for the cast, player and download workflow.