Photos that work in a two-person video

In this format the photo does one job: it tells the model who the person is. It is not a scene, not a background, not a pose. Everything else — the room, the camera, the performance — comes from the reference footage. That means the only thing your photo has to do well is show a face clearly.

It also means a beautiful photo can be a bad input. A heavily filtered portrait, a shot with the face in shadow, or a busy photo where the person is small in frame all give the render less to work with than a plain shot taken in daylight.

Four things, in rough order of how much difference they make:

Group photos are the most common one. It feels efficient — everyone is in the shot — but the model has to choose, and what it chooses is often a blend of two faces. Upload separately, one person at a time.

Screenshots and re-saved social photos are the second. Compression has already destroyed the fine detail, and the render amplifies what is left. If you have the original, use the original.

Wearing something you hate is the third, and it is easy to forget: wardrobe comes from the photo. If you want to see a particular outfit in the clip, upload a photo of that outfit.

Some photos are rejected, usually for a face that is too small, too soft, or turned too far away. When that happens the credits come back automatically, so the cost of a bad upload is a few minutes rather than money.

The useful response is to change the photo, not to press the button again. Re-sending the same image tends to produce the same refusal, because the reason was in the image.

You need permission from both people. This is not only a policy line — it is the difference between a clip that is funny and a clip that is a problem for whoever is in it. If you would not be comfortable with the person seeing the finished clip, the photo should not be uploaded.