All guides

How to Make a Tiny Desk AI Video from Photos

Published 1 November 2026

The short version: upload one portrait and you get back a vertical video of just over 19 seconds in which that face fronts a five-piece office band. Everything below is the long version, including the parts that go wrong.

Step 1: pick the photos

One adult per picture, head and shoulders, facing the camera, lit evenly, eyes visible. JPG, PNG or WebP, up to 10 MB each.

• Sharp beats flattering. An awkward photo that is in focus does better than a lovely blurry one.
• Skip group shots. Nothing here pulls one person out of a crowd; the whole picture is read as the person for that seat.
• Sunglasses, masks and heavy filters hide the parts that carry a likeness. Leave them out.
• Side profiles and faces tilted away give the weakest results.

Step 2: hand out the five seats

The band is a lead singer, someone on keys, someone on saxophone, and two backing singers, one left and one right. Each photo goes into one named seat, and the lead singer is the only seat you have to fill. Any seat you leave empty is played by a fictional portrait the site keeps for the purpose, so a single photo still produces a full band.

A fuller band costs nothing extra. One face or five, the price per video is the same.

Step 3: 480p or HD

Two sizes. The smaller one comes out at 480 by 854, which looks clean on a phone and is what most people post. HD comes out at 720 by 1280 and holds up better on a laptop screen or a television. The smaller size costs 10 credits per video and HD costs 15. If the clip is going into a group chat, we would not bother with HD; the extra sharpness disappears the moment someone watches it one-handed on a train.

Credits come in one-time packs: $12.99 for 10, $29.99 for 30, $49.99 for 60. There is no subscription.

Step 4: wait

We timed four videos at the smaller size and they came back in roughly four to six minutes each. HD takes a lot longer. The one HD run we have a stopwatch figure for took about half an hour, and a single measurement is not a promise, so plan for a long wait rather than a particular one. You can close the tab. The video waits in your account.

Step 5: what you download

Every video arrives as two files. One has the performance audio on it. The other is the same picture with no sound at all, which is the one you want if you are dropping the clip into an editing app or letting a social platform lay its own track over the top.

Both files are vertical 9:16, 24 frames a second, 457 frames, just over 19 seconds. We stepped through several finished videos frame by frame and found no watermark anywhere in them.

What makes it fail

Most of the trouble is in the photos.

• A face that is too small in the frame, too soft, or half covered.
• A file over 10 MB, or a format that is not JPG, PNG or WebP.
• A group photo dropped into a single seat.
• Now and then the video itself fails to come out. The credits go back to your balance automatically when that happens, so you do not have to write in and ask.

What it will not do

Everything you get is an AI reworking of one reference performance that never varies, so the list below is not a list of things we have not built yet. These are the shape of the product.

• The five camera angles and the moments the video cuts between them are always the same. You cannot reorder them, add one, or hold a shot longer.
• The music is fixed as well. There is no way to put a different song underneath before the video is made.
• Faces drift. You get something that looks like the person rather than a likeness you could swear to in court, and the movement and mouth shapes are approximate too. Nobody should treat the result as a reconstruction of the person in the photo.
• Music rights and the right to use somebody's face are two separate things, and this site hands you neither. If a real person is going into the band, get their permission. If you are posting somewhere with its own rules about music, read those rules.