Start here
Best photos for a Tiny Desk AI video
The model can only work with what the photo shows. These are the details that decide whether the person at the desk is recognisable or a stranger with their hair.
What a good input photo looks like
- One face, facing the camera. A three-quarter view is fine; a full profile is not.
- The face fills a good part of the frame. Crop a group shot down to one person before uploading.
- Even light. Window light beats a dark bar. Harsh side light hides half the face.
- Nothing over the face. No sunglasses, masks, caps pulled low, hands, or hair across the eyes.
- Sharp. Motion blur and a heavy beauty filter both destroy the detail the model needs.
- Plain clothing helps. Big logos and busy prints can bleed into the shot.
Why the close-up matters most
The clip pushes in on the lead singer for about six seconds. That is the shot people screenshot, so the lead vocals photo deserves the sharpest image you have. The backing seats appear in the wide shots, where slightly softer photos still work.
Photos of pets, cartoons and avatars
This product is built for people. A pet or an illustration will not map onto a human band member and usually produces a broken face. If you want a pet format, that is a different scene, not a different upload.
Old and scanned photos
A scanned team photo or a photo of a photo can work if the face is sharp and unobstructed. Expect the result to inherit the softness of the source: the model cannot invent detail that is not there.
Read next: the full generator walkthrough · why a face comes out wrong.