Updated 2026-11-11 · By Tiny Desk AI
Short answer: the best photo for Tiny Desk AI is a sharp, front-facing photo of one adult in soft, even light, framed so their hair and shoulders show. Skip group shots, tiny faces, heavy filters and sunglasses. The lead singer and the sax player get close-ups, so give those two seats your clearest photos.
Solo or five-person band · original sound
One photo per seat, one band
The video below is our own example, made with Seedance 2.5 from the five photos shown under it, one adult per seat.
What the best photo for Tiny Desk AI looks like
Tiny Desk AI doesn't paste a face onto the original singer. Each person is redrawn from their photo, including face, hair and clothes, inside the familiar desk-concert scene. The model can only keep what the photo shows, so a clear photo with plenty of detail does most of the work.
You don't need a stage photo, a microphone or a special outfit. An everyday phone photo against a plain wall is fine. What matters is that one face is large, sharp and evenly lit.
- One adult, alone in the frame, with no other faces, shoulders or hands at the edges.
- Facing the camera with both eyes visible. A slight turn is fine; a side profile isn't.
- Soft, even light on the face, ideally from a window in front of you.
- Sharp focus. Zoom in on the eyes: if they look soft or grainy, pick another photo.
- Framed from the top of the head to the chest or waist, so hair and shoulders show.
- No beauty filter, no sunglasses, nothing covering the mouth or chin.
Tiny Desk AI photo requirements: the fixed rules
Most of this page is advice. These points are rules.
If a photo is refused, the render stops and its credits return to your balance automatically. Replace it with an everyday photo of a consenting adult and start again.
- One photo per person. Each seat you fill takes its own photo; a single group photo can't be split into five people.
- Up to five photos per video: the lead singer plus up to four friends. Seats you leave empty keep their AI musician.
- JPG, PNG or WebP files, up to 10 MB each. A normal photo from your camera roll fits.
- Only you, or an adult who has agreed to be in the video.
- No minors, no celebrities and no public figures, including sports coaches and players.
Which photo to use for each seat
The video runs about 18 seconds and is built from four shots: the band (0 to 5.3 s), the lead singer's close-up (5.3 to 11.6 s), the sax close-up (11.6 to 15 s) and the band again to the end. The seats don't get the same camera, so they don't all need the same kind of photo. Left and right mean the sides of the frame as you watch it.
With one photo, you take the lead and the other four seats keep their AI musicians. Adding friends doesn't change the price: a solo video and a five-person video both use 10 credits in 480p or 20 in 720p HD.
- Lead singer: about 6 seconds in full close-up. This is the face people replay, and the camera is close enough to show every weakness in the photo. Use your sharpest, most front-facing photo, eyes open and nothing on the face. The person the video is about goes here.
- Front row at the keys: about 8 seconds across both band shots. Seated front and center, large in the frame, and on screen longer than the lead. Use a clear head-to-waist photo; the shirt or jacket in it will show.
- Back row left: about 8 seconds at medium size, standing behind the keys. Faces are smaller here, so hair and clothes do more of the recognizing. A bright jacket or a well-known hairstyle helps people spot who it is.
- Back row right: the same size and timing as the left. If two friends look alike, put one of them in a close-up seat instead of standing them side by side, so viewers can tell them apart.
- Sax player: about 3 seconds in close-up. It's short but large, and it lands as the surprise. Use another clear, front-facing photo, because a small or soft face shows here.
Common photos that give weak results, and why
When a finished video doesn't look like the person, the cause is usually the photo rather than the seat. These are the patterns worth avoiding, each with its fix.
- Group photos and crops of them. A face cropped out of a group is usually small and soft, and a stray shoulder or face at the edge gives the model something extra to copy. Pick a photo where the person is alone.
- Faces that are too small. A full-body shot from across the room leaves only a few pixels for the face. Move closer, or choose a head-to-waist photo.
- Heavy filters and beauty modes. Smoothing, face-slimming and big-eye effects change the face, and the video copies the filtered version, not the real one. Use the original camera photo.
- Sunglasses, masks and low hat brims. When the eyes or mouth are hidden, the model has to guess them. Show the full face.
- Dark or backlit photos. A bright window behind someone turns the face into a shadow. Turn around so the light falls on the face.
- Side profiles and big head tilts. The model needs the front of the face. Look at the lens.
- Blurry or re-saved images. Screenshots of screenshots, digital zoom and old compressed photos lose detail. Use the original file.
How to take a good photo in two minutes
If nothing on your phone passes, take a new photo. It's quicker than spending credits on a weak one.
- Stand facing a window in daylight. Avoid overhead spotlights and direct midday sun, which leave hard shadows under the eyes.
- Ask a friend to shoot from about 3 to 6 feet away at eye level, holding the phone upright.
- Frame from just above the head down to the waist, with a little space above the hair.
- Use the rear camera if you can; it's usually sharper than the front one. For a selfie, hold the phone at arm's length and at eye level.
- Turn off portrait blur, beauty mode and filters, and wipe the lens.
- Take off sunglasses and anything that hides the eyes.
- Wear what you want to wear in the band. Clothes and hair carry over from the photo.
- Keep a relaxed face. A neutral look or a small smile works; you don't need to pose as if singing.
- Take three or four shots and keep the sharpest one.
Consent and privacy: whose photo you can use
Only upload a photo of yourself or of an adult who has agreed to be in the video. Ask friends before you add them, and tell them it will be an AI video set to a real song. Sending them the result before you post is a simple courtesy.
Never upload photos of minors, including your own children. Celebrities and public figures aren't allowed either, and that includes the college football coaches the meme started with. Their faces aren't yours to put in a video.
When you post, turn on the platform's AI-generated label and don't present the clip as a real performance. Tiny Desk AI is an independent tool and isn't affiliated with NPR, Charlie Wilson or the Gap Band.
Your pre-upload checklist
Run through this before you press Make my video. It takes less than a minute per photo.
The finished video is a 9:16 MP4 of about 18 seconds with the original sound, plus a silent copy, and no watermark. A five-person HD video has recently taken about 15 minutes, and 480p is quicker. You can leave the page and come back to download it.
- One adult in the photo, nobody else at the edges.
- Face large, sharp and facing the camera, with both eyes visible.
- Even light on the face, with no strong shadow or backlight.
- Hair and shoulders inside the frame.
- No filter, no sunglasses, nothing over the mouth.
- JPG, PNG or WebP, under 10 MB.
- Your clearest photos on the lead singer and sax seats.
- Everyone in the photos is an adult who said yes.
What a good photo can't fix
A strong photo raises the odds, but AI video still has small glitches. Fingers on the keys can blur, a lip can land a beat off the song, and a hairline or jacket detail can shift when someone turns their head. 720p HD gives a sharper picture; it isn't a promise that every frame will be clean.
Watch the whole clip with sound before you post. If a render fails or is interrupted, its credits return automatically. Cash refunds follow the refund policy and need a message to support. The Single pack is $9.99 for 10 credits, one video in 480p, as a one-time purchase rather than a subscription.
Good photo vs weak photo, and why it matters
People in frame
- Good photo
- One adult, alone
- Weak photo
- Group shot, or a crop of one
- Why it matters
- Extra faces and shoulders give the model something else to copy
Face size
- Good photo
- Head to waist, face clearly large
- Weak photo
- Full body from across the room
- Why it matters
- A small face leaves little detail to rebuild from
Angle
- Good photo
- Facing the camera, both eyes visible
- Weak photo
- Side profile or strong tilt
- Why it matters
- The model has to guess the side it can't see
Light
- Good photo
- Soft daylight from the front
- Weak photo
- Backlit, dim or harsh overhead light
- Why it matters
- Shadows hide features, so they come out vaguer
Sharpness
- Good photo
- In focus, original file
- Weak photo
- Blurry, zoomed in or a screenshot
- Why it matters
- A soft photo gives a softer, less recognizable face
Filters
- Good photo
- Straight from the camera
- Weak photo
- Beauty mode, face-slimming or heavy filters
- Why it matters
- The video copies the filtered face, not yours
Eyes and mouth
- Good photo
- Fully visible
- Weak photo
- Sunglasses, mask or a low hat brim
- Why it matters
- Hidden features have to be invented
Hair and outfit
- Good photo
- In the frame, as you want them
- Weak photo
- Cropped off at the neck
- Why it matters
- Clothes and hair carry over into the band
The lead singer and sax seats show the face largest, so they need photos from the middle column most.
Frequently asked questions
What photo should I use for Tiny Desk AI?
The best photo for Tiny Desk AI is a recent, sharp photo of one adult facing the camera in soft light, framed from the head to the waist. A plain background is fine. Put your clearest photo on the lead singer, since that seat gets a 6-second close-up.
Can I use a selfie?
Yes, if it's sharp, front-facing and shows your hair and shoulders. Hold the phone at eye level and at arm's length, because a very close selfie can stretch the nose and crop the hair. A photo a friend takes from a few feet away usually works better.
Can I use one group photo for the whole band?
No. Each seat takes its own photo of one person. Faces cropped from a group shot also tend to be small and soft, so take a separate photo of each friend if you can.
Can I wear glasses or a hat?
Clear glasses are usually fine as long as your eyes are visible. Sunglasses aren't, because they hide the eyes. A hat is OK if the brim doesn't shade your face.
Can I upload a photo of my child or a celebrity?
No. Only upload yourself or adults who have agreed. Photos of minors, celebrities and public figures, including coaches, aren't allowed.
What if the video doesn't look like the person?
Check the photo against the checklist above and try a sharper, front-facing one; most likeness problems start with the photo. If a render fails or is interrupted, its credits return automatically. Cash refunds follow the refund policy and need a message to support.