DIY

Tiny Desk AI prompts

Want to make it yourself? These prompts work with AI video models that have an edit or reference mode. Render each shot, check it, then cut them together.

Before you start

Cut the source clip into the four shots below.
Pick one sharp, front-facing photo per person.
Render each shot with its prompt and the matching photos.
Re-render any shot where a face is off, then join them and lay the song back on.

Shot 1 · Band (0:00–0:05)

Replace the singers and the keyboard player in this video with the people in the reference photos, one person per position, keeping their faces, hair and clothes from the photos. Keep the camera move, lighting, desk, instruments and every mouth movement exactly as in the source video. Natural skin, no extra people, no text.

Shot 2 · Big laugh (0:05–0:12)

Replace the lead singer in this close-up with the person in the reference photo, keeping their face, hair and clothes. Keep the big open-mouthed laugh, the head movement, the microphone, the stage light and the timing exactly as in the source video. Photorealistic, no extra people, no text.

Shot 3 · Sax solo (0:12–0:15)

Replace the saxophone player with the person in the reference photo, keeping their face, hair and clothes. Keep the saxophone, the fingering, the body sway and the camera framing exactly as in the source video. Photorealistic, no text.

Shot 4 · Band again (0:15–0:18)

Same as shot 1: replace every visible performer with the matching reference photo, keep the motion, lighting and mouths from the source video. Match the faces used in shot 1 so each person looks the same across shots.

What to check in each render

The face holds when the head turns; side angles are where it slips first.
Mouths stay on the beat.
No original performer is left in any seat.
The same person looks the same across shots.

Rather skip the hour of editing? Try Tiny Desk AI does all four shots for 1 to 5 people in one render for $9.90. Compare the options.

Questions

Which AI video models can do this?

Models with an edit, video-to-video or reference mode, such as Kling and Seedance. Image-to-video alone won’t keep the original performance.

How much does doing it yourself cost?

One paid render per shot, more when a face comes out wrong, plus the editing time. Most people spend more than one video’s worth of credits on the first try.

Do I need the original clip?

Yes, as the source video for each shot. Use it only for a personal, non-commercial parody.

Skip the editing

Add a photo per seat and get the finished clip with the song.

Start the band →