Tiny Desk AI: Five Photos, One Desk Concert Video
Upload up to five portraits — one lead, up to four band positions — and Tiny Desk AI casts them into a single vertical desk-concert clip with the staging and camera cuts already built in. One photo is enough to start; the slots you leave empty stay exactly as they are.
Start my Tiny Desk AI videoUp to 5 faces · one video per run · about 15 seconds · 480p or 720p · sound included
1–5 photos · start with Lead (center) · empty slots keep the original band
How the five slots work
The frame has five places in it: one lead at the centre microphone, and four band positions around them. Every slot takes exactly one portrait, and every slot is optional — you fill them from the lead outwards and stop whenever you like.
Lead
Most screen time, and the face the camera returns to between cuts. This slot should get your cleanest photo, not your favourite one.
Backing left and backing right
Set back from the mic and sharing the frame with someone else. Faces here are never in close-up, so a slightly angled portrait is fine — and it's where the photos that weren't sharp enough for the lead can still live.
The close-up position
At some point this face fills most of the frame. Filters, soft focus and heavy brightness correction are all visible in that moment.
The fourth band position
Partly behind an instrument, so a normal head-and-shoulders crop works. Keep the face roughly the same size as the others or it will look like a different image was pasted in.
Upload photos in slot order — lead first, then the rest. The generator reads each file as a separate person and places it by slot, so the order you upload in becomes the order you see.
One, three or five faces — what each cast reads as
| Faces uploaded | What you see | Best for |
|---|---|---|
| 1 | A solo performance. One person alone at the desk, taking it far too seriously — which is the whole joke. | Your first test; a single-person joke |
| 2 | Two people sharing the mic; it reads as a duet, not a band | Couples, teammates, a running argument |
| 3 | The smallest cast that reads as a band | A clean trio beats a smeared five |
| 4 | Busy, one instrument position still empty-looking | When one person's photo isn't good enough |
| 5 | The full staging: lead plus four band positions | Teams, squads, a group-chat payoff |
The trade-off is simple: every face you add takes screen time away from the others. Five people at thumbnails that are already soft will read as five soft faces. If two of your five portraits are weak, upload three and let the rest of the frame stay as it is.
Which photos survive the crop
The model can only reproduce what it can actually read, and it fills every gap with a guess. Guesses are what make these clips look wrong.
| What you upload | What tends to happen |
|---|---|
| Straight-on phone photo, both eyes clear, even indoor light | Holds for the full clip |
| Sunglasses, low cap, scarf over the chin | Features drift, or blend into the face next to it |
| Side profile or head turned away | Cheekbones and jaw get invented, usually badly |
| A group photo cropped down to one head | Waxy and soft once scaled into the frame |
| Dim bar photo, heavy compression | Noise reads as skin texture |
| Screenshot of a photo on another screen | Banding, then blur |
The thumbnail test before you spend a run: shrink the photo until the whole face is the size of a fingernail. If you can still tell who it is, the model can read it. If not, reshoot — retouching the old file won't fix it.
Shoot all your portraits in one sitting, in the same room, under the same light. Mixed lighting is the fastest way to make a band look like five separate photos taped together.
Before you download: the three-check pass
A run that finishes is a run you've paid for, so review it before you walk away.
- Watch it end to end once, at full size. Don't judge a face from the first frame.
- Follow each face through the clip. Look for drift, feature swaps, and anyone who stops looking like themselves after a cut.
- Check the edges. No black bars, no cut-off shoulders, nobody pushed out of frame.
All three clear = download and post. Anything off = fix the photo, not the prompt, and run again.
When a face comes out wrong
| Symptom | Usual cause | Fix |
|---|---|---|
| A face dissolves halfway through | Motion blur or a soft file | Reshoot still; sharpening a blurry photo makes it look worse |
| Two people swap features | Similar faces, similar angles, similar light | Change one person's angle or the side of light in the reshoot |
| Black bars down the sides | Square or landscape source | Crop to vertical before uploading |
| Everyone looks like clay | Faces too small in the original frame | Get closer to your subjects |
| Only one position looks off | That's the weakest photo of the set | Replace it — the other slots are fine |
| A cut-off head or shoulder | Subject framed too tight, desk edge missing | Use a wider shot with shoulders and desk visible |
One rule that saves budget: the fix is almost always the source photo. Re-running the same weak file with a different idea produces another weak clip.
Running a series instead of a one-off
Every clip from this template shares staging, framing and cuts — only the cast changes. That makes a series cheap to produce and easy to keep consistent:
- Shoot everyone in a single session, one portrait each
- Name files by slot and person (lead-dana.jpg, band-left-sam.jpg) so you know what you're uploading
- Generate the lead slot alone first, check the face, then fill the remaining slots
- Keep a one-line log of which runs landed, so you don't regenerate what already worked
Rotate a new lead each week and you have a format, not a single video
Specs, limits and cost
| Item | Detail |
|---|---|
| Faces per video | 1 to 5, one portrait each, one video per run |
| Output | Vertical, desk-concert staging with camera cuts |
| Audio | Included in the export |
| Duration | About 15 seconds |
| Resolution | 480p or 720p |
| Price | Roughly $10 at 480p, roughly $20 at 720p — shown before you confirm |
| Failed runs | Failed runs are refunded |
Everything above is displayed before you confirm a run, so there's no surprise on the invoice.
Publishing rules
- Label it as AI-generated. TikTok, YouTube and Meta all have the toggle; it's the difference between a clip that stays up and one that gets reported.
- Only use photos you have the right to use — your own, or adults who agreed to appear. Public figures and press images aren't allowed.
- Don't present 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 describes the concert format the trend imitates; there's no affiliation with NPR.
FAQ
Can I make a Tiny Desk AI video with fewer than five photos?
Yes. One is enough to make a complete clip; each extra slot adds a person to the band.
What happens to the slots I leave empty?
They stay as they are in the template. Nothing is generated into them, and your clip is not marked incomplete.
Which of my photos should go in the lead slot?
The sharpest, straightest, best-lit one. The lead gets the most screen time, so a weak lead is the most visible mistake in the video.
Can the same person appear in two slots?
You can, but the routine gives each position a different part, so it reads as a duplicate rather than a band. Use different people, or leave the extra slot empty.
Why does one face look wrong while the others are fine?
Because the photos were shot under different conditions — different light, different angle, different distance. Reshoot that one portrait to match the rest.
Do I need to crop photos before uploading?
Only if they aren't vertical. A landscape or square photo gets padded, which shows up as bars at the sides.
Is this the official NPR Tiny Desk?
No. It's an independent AI video tool that recreates the crowded-desk staging people recognise. No affiliation with NPR or any artist.
Can I post the result commercially?
If the photos are yours and everyone in them has agreed, yes — with the platform's AI-generated label switched on. Don't imply it's real performance footage.
More concert-style AI templates
Try another crowd-moment template on AI Rap Reel.
Cast up to five faces in one Tiny Desk AI video — start with your best photo.
Make my Tiny Desk AI videoNot affiliated with NPR. "Tiny Desk" refers to the concert format this trend imitates. Independent AI tool by AI Rap Reel.