Tiny Desk AI Video — Make a Desk Concert Clip That Holds Up
One straight-on portrait is usually enough. Upload it, choose the slot it belongs in, and you get a vertical desk-performance clip with sound that you can post today.
Start a clip — one photo is enoughAbout 15 seconds · 9:16 · sound included · a first still-frame check before you pay
1–5 photos · start with Lead (center) · empty slots keep the original band
Good photos vs bad photos
Reads clean
Straight-on, both eyes clear, flat indoor light — the model can hold the face for the full clip.
Usually fails
- Sunglasses, cap pulled low, scarf over the chin
- Side profile or head turned away
- A group photo cropped down to one head
- Dim bar shot or heavy compression
- Screenshot of a photo on another screen
Sixty-second quick start
Step 1
Choose the slot
Choose the slot you're filling — centre, left, right, the close-up vocalist, or the sax player. You don't have to fill all of them; whatever you leave alone stays as it is.
Step 2
Upload one portrait
Upload one portrait for that slot. Front-facing, whole face visible, ordinary indoor light.
Step 3
Generate and review
Generate, then watch it end to end before you download. A take that finished is a take you paid for, so watch first.
If you're filling more than one slot, do it left to right from the centre: the model reads each photo as a separate person and placement follows the slot, not the order you dropped the files in.
What your photo decides
The model can only reproduce what it can read. Anything it can't see, it guesses — and guesses are what make these clips look wrong.
| What you uploaded | What tends to happen |
|---|---|
| Straight-on, both eyes clear, flat indoor light | Holds up for the full clip |
| Sunglasses, cap pulled low, scarf over the chin | Features drift, or blend into the person next to them |
| Side profile or head turned away | Cheekbones and jaw get invented, usually badly |
| A group photo cropped down to one head | Soft and waxy once scaled to 9:16 |
| Dim bar shot, heavy compression | Noise reads as skin texture |
| Screenshot of a photo on another screen | Banding, then blur |
The test before you spend a credit: shrink the photo to thumbnail size. If you can still see the whole face, the model can read it.
Casting slot by slot
Centre / lead
Most screen time, and the face the clip cuts back to. This slot gets your sharpest photo — if only one of your shots is clean, put it here.
Left and right backing
Half-turned, sharing light with someone else, and never in close-up. A slightly angled portrait works here, which is where you can use the photos that weren't good enough for the centre.
The close-up vocalist slot
At some point this face fills the frame. Soft focus, beauty filters and heavy HDR all show up at that moment.
The sax slot
The instrument covers part of the body, so a normal portrait crop is fine — just make sure the face itself isn't smaller than the others or it will read as a different resolution grade.
One person, three, five — what each cast reads as
A single upload is a solo performance: one person, alone at a desk, taking it seriously. That's the whole joke, and it needs nothing else.
Three uploads is where it starts reading as a band — enough bodies for the routine, still enough screen time for each face.
Five is the full cast. Every face gets less time on screen, so each photo has to be sharper than it would need to be in a two-person clip. If two of your five photos are weak, upload three people instead of five — a clean trio beats a smeared quintet.
Shoot all your portraits in one sitting, in the same room and light. Mixed lighting is the fastest way to make a group look pasted together.
Making a series instead of one clip
Clips in this format stack: the same staging, the same framing, the same song, only the cast changes. That makes a series cheap to produce — one shoot, one slot at a time, a new lead each week, the same file naming so you can tell takes apart.
A workflow that keeps it orderly: photograph everyone in one session → name files by slot and person → generate the centre slot first → only after it looks right, fill the remaining slots → keep a note of which takes landed, so you don't regenerate the ones that already looked fine.
When it comes out wrong
| Symptom | Usual cause | What to do |
|---|---|---|
| A face dissolves halfway through | Motion blur in the source photo | Reshoot still, don't sharpen the old file |
| Two people swap features | Similar faces, similar angles, similar light | Change one person's angle or light in the reshoot |
| Black bars down the sides | Square or landscape photo | Crop to vertical before uploading |
| The desk edge is missing | Subject too tight in frame | Use a wider shot with shoulders and desk visible |
| Everyone looks like clay at 9:16 | Faces small in the original frame | Take the photo closer |
| Only one slot looks off | That photo is the weakest of the set | Replace it; the other slots are fine |
One rule that saves money: a clip that completes is charged, even if a face is off. Fix the photo before you re-run, and use a short preview take when you're unsure about a face.
Specs and cost, in plain terms
| Item | Detail |
|---|---|
| Length | About 15 seconds — one performance section, not a montage |
| Frame | 9:16 vertical, built for TikTok, Reels and Shorts |
| Audio | Included in the export |
| Resolution | Pick 480p or 720p before you generate |
| Cost | Roughly $10 at 480p, roughly $20 at 720p, at the $9.99 / 850-credit pack rate |
| Packs | From $4.99 |
| Failures | Failed runs are refunded |
| Preview | A short low-res take costs less — the cheap way to test a face |
Credits and an approximate dollar figure are shown before you confirm, so nothing is a surprise after the fact.
Before you publish
- Turn on the platform's AI-generated content label. TikTok, YouTube and Meta all have the toggle, and it's the difference between a clip that stays up and one that gets reported.
- Use your own photo, or a photo of adults who agreed to appear. Public figures and press images aren't allowed.
- Don't caption it as real footage of the artist, the band or the show.
- Keep it vertical and keep the sound — the format depends on both.
- This is an independent tool. "Tiny Desk" here refers to the concert format the trend imitates; there's no affiliation with NPR.
FAQ
Do I have to fill all five slots?
No. Empty slots stay exactly as they are, so a one-photo clip is a complete clip.
What photo should I start with?
A phone photo where the face is sharp and lit evenly from the front. Straight out of the camera beats anything that has been through a filter, a screenshot, or a messaging app's compression.
Can I use the same person in two slots?
You can, but the routine gives each slot a different part, so it reads as a duplicate rather than a band. Use different people, or leave the extra slots empty.
Someone in the photo won't give permission.
Then don't use it. Also skip anything you found online, and anything showing a public figure — those get rejected and the run may still count.
Can I post this from a brand or work account?
Yes, if it's your own content and you label it as AI-generated where the platform asks.
What if the generation never finishes?
Reserved credits come back automatically when a job fails on our side. Check your creations list for the status before retrying.
Can I do this on my phone?
Yes. The uploader runs in a mobile browser, no app install.
Can I ask for a different song or a different scene?
Not on this page — this one is a fixed routine. The main generator takes text prompts if you want a different desk scene.
More concert-style AI templates
Try another crowd-moment template on AI Rap Reel.
Make a Tiny Desk AI video from one photo
Start a clip — one photo is enoughNot affiliated with NPR. "Tiny Desk" refers to the concert format this trend imitates. Independent AI tool by AI Rap Reel.