Tiny Desk AI prompts: scenes vs fixed video templates

A starting prompt for an original desk-concert scene

“Create a vertical desk-concert scene in a cozy rehearsal room, with bookshelves, warm practical lighting and a small band behind a desk. Use the supplied portrait for the lead performer, keeping their face, hairstyle and clothing recognizable. Frame the lead from the waist up at a microphone, with a keyboard player and backing musicians clearly separated in the background. Keep the camera steady and avoid text or logos.”

This is a writing starting point, not a tested preset or a promise of an exact Outstanding remake. Use it only in a tool that accepts the relevant reference images. Adapt framing, movement and duration to that tool’s supported controls rather than assuming the prompt can override them.

How should I describe a multi-person cast?

List each reference image and its role before describing the scene: portrait A is the lead, portrait B is the keyboard player, and portrait C is a backing singer. Specify who should stay unchanged. Avoid describing several people only as “the singer,” which leaves the intended mapping unclear.

A written cast list is useful for planning, but the model must actually support multiple identity references. Adding five names to a single-reference workflow does not create five-person support. Check a full example from the same mode you plan to buy.

Can a prompt reproduce the Outstanding song and lip-sync?

Do not assume so. Audio generation, reference-audio retention and lip-sync are separate capabilities. Ask whether the download includes newly generated music, a supplied track or no sound. Check the actual exported file rather than relying on music playing over a website demo.

For a recognizable performance remake, the reference clip and timing may matter more than adding more adjectives to the prompt. Using the name of a song does not establish permission to distribute its recording. Use audio you are entitled to include or add an available licensed track in your publishing platform.

How do I improve a result without changing everything?

Compare the same portrait and role assignment first. Change one input at a time: photo, framing, reference clip or prompt. Keep the original output and record what changed. Check face identity, hands, instrument overlap, camera cuts and audio separately.

If only the sax shot fails while the lead looks correct, report that shot instead of treating the whole idea as a bad portrait. Repeated full-video retries can cost more without revealing which input is responsible.

Do I need to paste a prompt into this website?

The current cast planner uses photo and role choices; it is not a general prompt editor. This guide helps you understand the workflow and compare tools. Follow the creator’s available controls when choosing a template.

A little context before you create