Short answer
Use one clear half-body photo per seat: head and shoulders in frame, face square to the camera, even light from the front, no sunglasses. The clip is a medium shot of two people from the chest up, so a tight headshot crops badly and a full-body photo makes the faces too small to keep. One photo per person, never two people in one photo.
Try it with two photos58th Floor AI photo guide
Two seats, one medium shot, fifteen seconds — so the photo is doing more work than usual.
Updated 2026-11-03
The short checklist
- One person per photo. Two people in one frame fuse into one face.
- Head and shoulders inside the frame, with a little room above the hair.
- Face square to the lens, eyes visible, no sunglasses and no hat brim over the eyes.
- Even light on the face — a window or a lamp in front of you, never behind.
- A plain background is ideal, but the render drops it anyway; a busy room mostly costs sharpness.
Per seat
The left seat is the hook. It opens the clip and stays in frame the longest, so this is the photo to spend time on: sharper, brighter and more front-facing than the second one.
The right seat is the verse. It enters later and turns half a step toward the microphone, so a straight-on photo with the chin slightly up keeps the face readable during the turn.
What actually breaks
| Photo | What happens |
|---|---|
| Two people in one frame | The model merges them into one face in one seat |
| Sunglasses | Eyes are invented, and the face stops reading as the same person |
| A screenshot of a screenshot | Soft edges and colour banding survive the render |
| Strong side light or backlight | One half of the face goes dark and stays dark |
| Full-body shot | The face lands too small in a chest-up frame |
Casting two people
Upload the taller person on the left if you are unsure: the left performer stands slightly forward in the reference, and a taller left figure matches the blocking better. Anything close works — the model rescales each person to the seat rather than keeping their real height difference.
Both seats are independent. Adding a second photo never changes the price, and leaving one empty keeps the original performer instead of cropping the frame.
More on portraits in general: photo requirements. Ready to render: 58th Floor AI.