Best Photos for a 5-Person Tiny Desk AI Video

One person in each of five photos

Prepare one portrait for lead vocals, backing left, backing right, keys and vocals, and saxophone. Every role requires an upload. A single group photograph is not a substitute for those five inputs, even if the group contains exactly five people.

The model uses each image as an appearance reference. When two faces share an input, it may choose the wrong face or blend details. If your only source is a group photograph, make separate crops that show just one person each, with enough real detail to recognize the face. A tiny, blurry crop does not become detailed simply because it is enlarged.

Match the photo to the role

The lead is shown in a smile close-up, so prioritize a clear view of the eyes, mouth and hair. Keys and the backing singers appear together and return in other shots; distinct hair and clothing help make it easier to review whether the right person remains in the right role. The saxophone role needs a recognizable face and enough upper-body detail to guide its appearance.

These are preparation tips, not guarantees. A photograph cannot force the model to preserve a person across every frame. See what to expect from face consistency and lip-sync.

Show the face and the outfit

A well-lit, front-facing or lightly turned head-to-waist photograph is a useful starting point. Avoid strong shadows across the eyes, sunglasses, hats obscuring the hair, hands over the mouth, or extreme filters. Use a view where the face is a meaningful part of the frame rather than a tiny detail.

If you want a recognizable jacket or hairstyle, include it. A tight headshot gives little information about body shape or clothing below the shoulders, and a request for an unseen outfit leaves more for the model to invent. Fine logos and lettering are not reliable identity anchors; their shape or spelling can change.

Mirror selfies can carry a phone or camera into the generated scene. Busy backgrounds can also influence what the model adds. Prefer a simple photograph with no prominent object hiding the body, rather than trying to correct a complicated input entirely with text.

Files and upload checks

The upload accepts JPEG, PNG and WebP files up to 10MB each. The application resizes and compresses images before storage. Use the original good-quality photograph where possible; repeatedly saving a heavily compressed screenshot can remove facial detail before the upload process even starts.

Check every thumbnail after upload. Confirm that the lead slot contains your intended lead and that left and right backing roles are not swapped. Remove and replace an incorrect image before generation. All five roles must be populated to enable the task.

A short pre-generation checklist

1. Exactly one visible person in each image.
2. Eyes, mouth and hair are visible without a covering object.
3. Useful upper-body and clothing detail is included.
4. The correct person is assigned to each role.
5. Everyone whose photo you use has appropriate permission.
6. You have reviewed the selected quality and credit cost.

No photo preparation method guarantees perfect identity, clothing or mouth timing. Better inputs reduce ambiguity, but the output still needs review. Continue to the step-by-step guide or open the studio.