How to Make the Outstanding Tiny Desk AI Video

Updated October 28, 2026 · Product walkthrough

Your group chat already has a lead singer. It is probably the friend who talks through every meeting, celebrates a fantasy win too early, or insists they know the song. An Outstanding Tiny Desk AI video turns that casting joke into a short desk-concert performance. You bring the portraits; the generator makes the scene and motion.

This guide follows the current product: one to five photos, one person per photo, and a 15-second vertical MP4. You do not need a video editor or a written prompt. Start with the people and the joke, then work through the four steps below. Generation time varies, so leave room to review the result before you post it.

Your cast: five available positions
Backing vocal left
Backing vocal right
Keys
Sax
Lead · front center
Small desk + microphone

Role map, not a screenshot or a guaranteed camera frame. Unfilled positions use fictional musicians.

What you need

Have a clear portrait of each participating adult, permission to use those images, and an account for saving your videos. The uploader accepts JPG, PNG, and WEBP files up to 8 MB each. A phone portrait is fine; sharp facial detail matters more than a dramatic background.

Budget one credit for each generation. A cast of five uses the same one credit as a solo performance; the pack size counts videos, not people. Credits are bought in one-time packs and are valid for 12 months. There is no subscription and no free retry. Check the current packs below before committing to a render.

For example, three friends can share a three-person performance while using only one video credit. Buying a three-video pack gives you three generations; it does not unlock three photo slots. All cast sizes from one to five are available for each generation. Decide whether you want one shared clip or separate clips before choosing a pack.

Step 1: Cast the band

Choose a lead first. This is the front performer, so give the role to the person the caption is about. For an office joke, that might be the colleague presenting a very optimistic deadline. Add keys and sax for a three-person group, or fill the two backing-vocal positions for five people.

Each person needs a different position. The available choices are lead, keys, sax, backing vocal left, and backing vocal right. The lead carries the main vocal; the instrumentalists play their instruments, while the backing singers support the performance. You cannot put two people into the same position.

One person is enough to start. Empty positions are filled with fictional musicians rather than copies of your photo. If you are comparing possible leads, decide before generating: changing the cast after a completed video means starting another generation. Keep the joke friendly and get agreement from everyone whose face you use.

Homepage uploader with one lead photo, band role selection, scene and soundtrack controls
Actual uploader with the existing demo portrait. Roles, scene, and audio are reviewed before generating.

Step 2: Pick and crop the photos

Use a separate image for every person. A group selfie is not five reference photos: the uploader maps one image to one role. Ask friends for their own portraits rather than cutting several tiny faces out of a distant crowd. A face that is already small or blurred will not gain detail simply because you crop it.

Open the homepage uploader, add a photo, and confirm its 9:16 crop. Keep the entire face visible, with room around the hair and chin. Repeat for the other people, then check each role dropdown. Changing to an occupied role swaps the positions, so review the whole lineup after making a change.

A useful comparison is a well-lit portrait at eye level versus a dark side profile with sunglasses. Choose the first. Avoid masks, strong beauty filters, blown-out windows behind the head, or hands covering the mouth. Use the photo you would pick to help a stranger recognize that person, rather than the one with the most elaborate setting.

You can prepare the photos on a phone or on a computer. If you use the phone-upload handoff, review the resulting crop and role on the generator before you submit. Do not assume that the order friends sent their pictures in is the same as the final band assignment. A quick role check prevents an avoidable second attempt.

The 9:16 portrait crop interface with zoom and Use this photo controls
Actual photo-crop screen using the existing demo portrait. Keep the whole face visible before confirming.

Open the homepage uploader →

Step 3: Generate the video

Before submitting, review every face, crop, and position. Choose a scene such as Office or Garage, then the supplied Original audio reference or the AI-generated audio alternative. Original uses the supplied “Outstanding” clip as a reference; choosing it does not turn your upload into an official artist performance or provide blanket music clearance for every use.

Confirm that you have permission for every portrait and that all participants are adults. Adding another photo resets consent, so complete the lineup before checking the box. If you need to sign in or buy credits, follow the prompts and return to the generator. When there are no credits, the purchase chooser offers the available packs.

The service first prepares the scene and then generates the moving performance. Keep the job page available and check its status instead of repeatedly submitting the same request. The output is generative, so exact facial detail, instrument movement, and lip sync can vary. The format is fixed, but the model does not guarantee a particular sequence of camera cuts.

A failed render returns its credit automatically. A completed video that you would like to change is different: there is no free retry, and another generation uses another credit. If a job appears stuck, check My videos before starting again, and use the site feedback entry if you need help explaining what happened.

Homepage uploader with one lead photo, band role selection, scene and soundtrack controls
Actual uploader with the existing demo portrait. Roles, scene, and audio are reviewed before generating.

Step 4: Download and post it

When the job succeeds, download the MP4 from My videos and watch it once with sound. Check that each face is recognizable, the lead is the person you intended, and the joke still works without a long explanation. Save a local copy before switching devices or composing a social post.

The 9:16 format fits vertical posts on TikTok, Reels, and Shorts. A short caption usually does more work than a list of tags: “The team after calling that meeting a quick sync” gives viewers the setup immediately. Another option is “Our fantasy league after a perfectly normal Sunday.” Adapt the wording to people who have agreed to appear.

Identify the clip as AI-generated and use any disclosure tools required by the platform. Tags such as #AIVideo and #OutstandingTinyDesk can describe it; they do not guarantee reach. Do not imply an NPR endorsement or upload it as a real concert recording. If the destination flags music, review that platform’s rules and the rights needed for your use.

Existing example of a generated desk-concert performance
Existing generated example, not a new render made for this tutorial. Review your own MP4 with sound before posting.

Fixing a face that looks off

First check the reference rather than trying to solve everything with another paid render. Is the face obscured, heavily filtered, or much smaller than the other portraits? Replace that photo with a clear, evenly lit image. Then confirm that each role is attached to the intended person.

For a smaller first experiment, use one lead or a three-person cast whose faces are visually distinct. That is a way to simplify your own review, not a guarantee of a better model result. Do not keep buying attempts without changing an identifiable problem. Model rejection, a technical failure, and an accepted result you dislike have different consequences; read the terms before expecting a cash refund.

Choose a video pack

One credit per generation · Valid for 12 months · No free retry. See all pricing details.

Frequently asked questions

Can I make it with one photo?
Yes. Add one adult portrait as the lead. Fictional musicians fill the supporting scene; you do not need to supply five people.
Can I upload one group photo?
Use one person per photo. Prepare separate portraits and assign one unique role to each upload.
Do five people cost five credits?
No. One generation costs one credit whether you upload one photo or five. Packs count completed video attempts, not cast members.
How long does generation take?
There is no guaranteed completion time. Model queues and processing vary; check the job status in My videos rather than expecting a fixed three-minute turnaround.
Can I redo a completed video for free?
No. There is no free retry. A technical render failure returns its credit automatically; a new generation after a completed result uses another credit.
Do I need CapCut?
No. The product generates the clip without a timeline editor. You can optionally edit the downloaded MP4 elsewhere before posting.