A Tiny Desk AI video prompt should identify the source clip, map each reference photo to a specific performer and say which roles must stay unchanged. “Make a Tiny Desk video of my friends” leaves those decisions to the editor.
The example below uses the same prompt builder as the EasyVideoAI Tiny Desk workflow, showing all five selected roles. In the studio, the prompt is built automatically from the roles you actually fill. There is no custom prompt field; individual results still need review.
What to put in a Tiny Desk AI video prompt
Start by separating a source-based edit from a newly generated desk concert. If you want to recast an existing performance, name the supplied clip as the source. Otherwise an editor may interpret the request as an entirely new room, band and camera sequence.
Then write a role map. Describe the lead singer, keyboard player, two backing singers and saxophonist individually. A list of five names is not enough if the editor has no reliable connection between those names, the reference images and the people in the video.
Use labels that remain meaningful across cuts. “Viewer-left backing singer in the band shots” is more specific than “the person on the left,” which can refer to someone else after the camera changes. In this source, the lead wears a navy visor and star jacket, the keyboard player a green polo, the left backing singer a blue shirt and white-and-blue cap, and the right backing singer a maroon polo with yellow trim. The saxophonist wears a dark jacket over a pale blue shirt. These clothes identify the original performers. The selected reference supplies the replacement person’s appearance and outfit. Keep each assignment attached to the same person when the camera cuts.
An example Tiny Desk AI editing prompt
Replace each bracketed reference with the input label used by your chosen editor. Keep unselected roles marked KEEP ORIGINAL. These labels describe the plan; they are not universal API field names or upload syntax.
[tiny-desk-v4-english-full-person]
Edit the existing complete performance in @Video 1, replacing only the selected original performers listed below. Preserve the source performance, instruments, setting and editing. Complete each assigned replacement from the FIRST frame in which that performer appears through their LAST visible frame. Selected backing singers and partially visible performers are required subjects too; do not overlook them because they are farther from the camera.
WHOLE-PERSON REPLACEMENT (highest priority): Replace each selected original performer COMPLETELY with the exact person depicted in their assigned image. That reference person must become the performer in the video: their face, entire head, hairstyle, skin, physique, body proportions, clothing and personal accessories. Every visible human body part and item of clothing must belong to that same reference person. Replace the whole person, not a set of borrowed characteristics, a similar-looking person, or a new face attached to the original person's body. The original selected person's head, hair, body appearance and costume must be fully replaced from their first visible frame. Retarget the original performance to the reference person.
REFERENCE AUTHORITY: Clothing, headwear, screen position and instruments described below identify people in the ORIGINAL video only. These are tracking cues, NOT wardrobe instructions for the replacements. The assigned image shows WHO must replace that person, with their complete appearance and outfit. The source video supplies only that role's motion, blocking, performance, camera and scene; it does not supply the selected replacement's human appearance or clothing. An unselected person keeps their original appearance AND clothes. Image numbers refer to the actual uploaded reference array; photo slots label the UI and do not create extra references.
FIXED PERSON-TO-PHOTO ASSIGNMENTS (highest priority):
REPLACE ONLY Lead singer (photo slot 1) [lead]: ORIGINAL TARGET: navy visor and navy star-emblem zip-up jacket; the opening and central close-up performer. Replace this ENTIRE original performer with the EXACT PERSON shown in @Image 1. Put that person, with their own face, hairstyle, physique, complete clothing and personal accessories, into this role for every appearance. Never assign this person to another role.
REPLACE ONLY Keyboard player (photo slot 2) [keyboard]: ORIGINAL TARGET: green polo and black headset; seated at the red keyboard, also partly visible at the lower right of the saxophone shot. Replace this ENTIRE original performer with the EXACT PERSON shown in @Image 2. Put that person, with their own face, hairstyle, physique, complete clothing and personal accessories, into this role for every appearance. Never assign this person to another role.
REPLACE ONLY Left backing singer (photo slot 3) [backing-left]: ORIGINAL TARGET: blue shirt, white-and-blue cap and black headset; standing behind the keyboard on the viewer's left. Replace this ENTIRE original performer with the EXACT PERSON shown in @Image 3. Put that person, with their own face, hairstyle, physique, complete clothing and personal accessories, into this role for every appearance. Never assign this person to another role.
REPLACE ONLY Right backing singer (photo slot 4) [backing-right]: ORIGINAL TARGET: maroon polo with yellow trim and black headset; standing behind the keyboard on the viewer's right. Replace this ENTIRE original performer with the EXACT PERSON shown in @Image 4. Put that person, with their own face, hairstyle, physique, complete clothing and personal accessories, into this role for every appearance. Never assign this person to another role.
REPLACE ONLY Saxophonist (photo slot 5) [saxophone]: ORIGINAL TARGET: gray hair, dark suit jacket, pale blue shirt and black headphones; the performer playing the gold saxophone. Replace this ENTIRE original performer with the EXACT PERSON shown in @Image 5. Put that person, with their own face, hairstyle, physique, complete clothing and personal accessories, into this role for every appearance. Never assign this person to another role.
SHOT-BY-SHOT BINDINGS: Times are relative to the source start and locate its existing shots. Follow the source's actual cuts and frame timing rather than creating new cuts at rounded timestamps.
00:00.000–00:01.333 Opening side-profile close-up of the lead singer. Lead singer [lead] → replace the ENTIRE performer with the EXACT person shown in @Image 1, including their complete appearance, hairstyle, physique and outfit. Track the original person identified by: navy visor and navy star-emblem zip-up jacket; the opening and central close-up performer. Apply all selected replacements in this frame SIMULTANEOUSLY; leave unselected people unchanged. Distance, a side profile, a cap brim or partial occlusion must not cause a selected role to be skipped.
00:01.333–00:06.708 Three-person band shot: keyboard player in front, viewer-left backing singer behind, viewer-right backing singer behind. Keyboard player [keyboard] → replace the ENTIRE performer with the EXACT person shown in @Image 2, including their complete appearance, hairstyle, physique and outfit. Track the original person identified by: green polo and black headset; seated at the red keyboard, also partly visible at the lower right of the saxophone shot; Left backing singer [backing-left] → replace the ENTIRE performer with the EXACT person shown in @Image 3, including their complete appearance, hairstyle, physique and outfit. Track the original person identified by: blue shirt, white-and-blue cap and black headset; standing behind the keyboard on the viewer's left; Right backing singer [backing-right] → replace the ENTIRE performer with the EXACT person shown in @Image 4, including their complete appearance, hairstyle, physique and outfit. Track the original person identified by: maroon polo with yellow trim and black headset; standing behind the keyboard on the viewer's right. Apply all selected replacements in this frame SIMULTANEOUSLY; leave unselected people unchanged. Distance, a side profile, a cap brim or partial occlusion must not cause a selected role to be skipped.
00:06.708–00:13.000 Continuous close-up of the SAME lead singer seen in the opening. Lead singer [lead] → replace the ENTIRE performer with the EXACT person shown in @Image 1, including their complete appearance, hairstyle, physique and outfit. Track the original person identified by: navy visor and navy star-emblem zip-up jacket; the opening and central close-up performer. Apply all selected replacements in this frame SIMULTANEOUSLY; leave unselected people unchanged. Distance, a side profile, a cap brim or partial occlusion must not cause a selected role to be skipped.
00:13.000–00:16.417 Saxophone solo; the keyboard player's partial profile is ALSO visible at the lower right. These are TWO DIFFERENT performers. Saxophonist [saxophone] → replace the ENTIRE performer with the EXACT person shown in @Image 5, including their complete appearance, hairstyle, physique and outfit. Track the original person identified by: gray hair, dark suit jacket, pale blue shirt and black headphones; the performer playing the gold saxophone; Keyboard player [keyboard] → replace the ENTIRE performer with the EXACT person shown in @Image 2, including their complete appearance, hairstyle, physique and outfit. Track the original person identified by: green polo and black headset; seated at the red keyboard, also partly visible at the lower right of the saxophone shot. Apply all selected replacements in this frame SIMULTANEOUSLY; leave unselected people unchanged. Distance, a side profile, a cap brim or partial occlusion must not cause a selected role to be skipped.
00:16.417–00:19.041667 Return to the SAME keyboard player and the SAME two backing singers from the first three-person band shot. Keyboard player [keyboard] → replace the ENTIRE performer with the EXACT person shown in @Image 2, including their complete appearance, hairstyle, physique and outfit. Track the original person identified by: green polo and black headset; seated at the red keyboard, also partly visible at the lower right of the saxophone shot; Left backing singer [backing-left] → replace the ENTIRE performer with the EXACT person shown in @Image 3, including their complete appearance, hairstyle, physique and outfit. Track the original person identified by: blue shirt, white-and-blue cap and black headset; standing behind the keyboard on the viewer's left; Right backing singer [backing-right] → replace the ENTIRE performer with the EXACT person shown in @Image 4, including their complete appearance, hairstyle, physique and outfit. Track the original person identified by: maroon polo with yellow trim and black headset; standing behind the keyboard on the viewer's right. Apply all selected replacements in this frame SIMULTANEOUSLY; leave unselected people unchanged. Distance, a side profile, a cap brim or partial occlusion must not cause a selected role to be skipped.
CROSS-CUT IDENTITY LOCK: The green-polo/red-keyboard original performer is keyboard in BOTH band shots and in the lower-right partial profile during the saxophone solo. The gray-haired/suited/gold-saxophone original performer is saxophone, a DIFFERENT person. Keep their assigned identities separate while both are visible. On the final cut back to the band, keyboard must retain the SAME identity and outfit used in the first band shot; NEVER switch keyboard to the saxophone reference. The blue-shirt/white-blue-cap performer and the maroon-polo/yellow-trim performer keep their own separate bindings in BOTH band shots. Follow each original person's track, not whichever face occupies a screen position. Do not insert a performer into a shot where they were absent.
THE EXACT REFERENCE PERSON, HEAD TO TOE: Preserve the assigned person's recognizable identity and full appearance, including their face, head shape, hairstyle, hair color, skin tone, age appearance, physique and visible body proportions. Their neck, torso, arms, hands and any visible legs must read as parts of this SAME person, not the original performer's body beneath a changed face. Dress them in their own complete visible outfit from that SAME image: shirt, outerwear, neckline, sleeves, cut, fit, colors, materials, lower garments, footwear and personal accessories wherever shown. Do not preserve or combine the original person's clothing, body appearance or hair with the reference person. Remove original non-performance headwear unless it is also present in the assigned image. Functional performance headsets, microphones and instruments remain scene props and must fit the reference person naturally. For body or clothing areas not visible in the photo, use minimal consistent completion without inventing a new character or outfit. Retain the video's motion and camera perspective rather than copying the photo's static pose or background. Keep this exact same person and complete outfit across every cut, head turn, occlusion and reappearance; no morphing, delayed replacement, costume switching or reversion to the original performer.
PERFORMANCE AND SCENE: Each replacement follows their own original performer's movement, position, gestures, expressions and mouth timing. Keyboard stays at the same red keyboard; saxophone stays in the solo position playing the same gold saxophone, with natural mouthpiece contact and finger placement. Each backing singer follows their own original gestures. Preserve people/prop occlusion and depth order; do not reveal hidden body parts. Keep the bookshelves, objects, microphones, camera positions, camera movement, framing, lighting and depth of field. Preserve all other people. Do not add, duplicate, merge, exchange or remove performers, or add text, split screens or new transitions.
DELIVERY: Replace exactly 5 assigned role(s); preserve the 0 unassigned role(s). Empty slots always keep the original person. Never shift assignments to fill gaps or assign photos in shot-appearance order. Keep the original 9:16 composition and complete 19.041667-second picture, from the original first frame to the original last frame, without extension, padding or speed changes. The input is silent. Output silent video only, without generated music, dialogue or sound effects; the application restores the original audio afterward.A prompt can express the intention, but it cannot add missing capabilities to an editor. Check whether the tool supports source-video editing, multiple references and stable role assignments before using the brief. For a template workflow, the interface may handle those instructions for you.
Adapt the prompt for one person or two friends
For a saxophone-only edit, put the saxophonist’s reference in slot 5 and leave the first four roles marked KEEP ORIGINAL. Do not shorten the role map into a one-item list labeled “person 1”; that loses the connection to the fixed saxophone slot.
For two friends on vocals and sax, fill slots 1 and 5. Explicitly preserve the keyboard player and both backing singers. If your chosen roles are different, change the role map rather than adding a vague sentence such as “put my friend somewhere in the band.”
For a full band, provide a distinct reference for every selected role and check the labels before submission. The friends casting guide has a slot table and example band plans.
Keep visual instructions and audio handling separate
Changing a face is a visual task. Keeping the soundtrack is a separate requirement. State that you want the source audio and timing preserved, but confirm how your editor handles audio rather than relying on a sentence in the prompt.
If visual processing produces a silent result, the workflow needs a separate restoration step using the source soundtrack. Check the opening, the camera cuts and the ending after combining them. Matching a portrait does not supply a new singing voice, and a request to preserve sound does not guarantee word-for-word lip sync.
In the supplied Tiny Desk material, the picture is about 19 seconds and the clip already contains audio. The homepage lets you watch and listen to that source; assign photos and review the quote before submitting a personalized performance.
What to change when an assignment is unclear
If a face appears in the wrong role, first check the input labels. Repeating “keep the faces accurate” will not fix a reference that was assigned to the wrong performer. Keep a complete five-role map even when you select only one person.
If two identities blend together, use separate, clear portraits and simplify the brief. Avoid contradictory instructions such as preserving the original face while replacing it with a new reference. Review the photo preparation advice before rewriting a long prompt.
For an unselected person who changes, make the preservation rule explicit and review the editor’s masking or role-selection controls. Text instructions alone may not enforce that boundary. Inspect the actual result before sharing it.
Tiny Desk AI prompt questions
Where do I paste this prompt on Tiny Desk AI?
The studio has no free-text prompt box. Its labeled photo slots create the role assignments automatically. The example shows the all-five-role version; empty slots are preserved in the actual request.
Is there one official prompt that always works?
No prompt on this page has been validated as a universal preset. Results depend on the editor, the source clip, input support and reference quality. The brief is a starting point for making those choices explicit.
Can I write KEEP ORIGINAL instead of supplying every photo?
That is how this brief describes an unselected role. In the Tiny Desk portrait layout, leave that slot empty. Another editor may require a mask or a different setting to preserve a performer, so check its controls.
Use the Tiny Desk AI video prompt as a clear role plan, then follow the full workflow to prepare photos, review the source and check what your chosen tool actually supports.