Tiny Desk AI video fixes: the face, the laugh, the sax, the sound
Most broken Tiny Desk AI videos come from three things: the photo, the model splitting a 19-second video in two, and the platform you post to. This page lists each problem with its cause and the fix that works on this generator.
Problem, cause, fix
| Problem | Why it happens | Fix |
|---|---|---|
| The singer's face changes halfway through | Models that stop at 15 s make a 19 s video in two parts, cut at 0:13; a soft or side-on photo gives each part room to guess. | Use a sharp, front-facing photo. For one continuous pass, pick Seedance 2.5, which makes all 19 s at once. |
| The face warps in the laugh close-up | The laugh is the biggest expression in the clip; a neutral, closed-mouth photo leaves the model inventing teeth and cheeks. | Upload a photo where they smile with teeth showing, eyes visible, head and shoulders in frame, even light. |
| The sax player gets the keys player's face | A group photo, or two similar photos, gives the model two faces to mix; the more seats you fill, the more faces it has to keep apart. | Use band mode with one person per photo in its own seat (keys, left and right backup, sax). Never upload a group shot; if faces still swap, leave the backup seats empty so AI performers play them. |
| Lips don't follow "Outstanding" | The motion reference carries head movement, not mouth shapes; the song is added after the video is made. | Keep the original audio (it is already in sync with the cuts). The higher models, Seedance 2.0 Pro and Seedance 2.5, usually follow the head motion more closely, at a higher price. |
| 480p looks awful after uploading to TikTok | TikTok recompresses everything; a 480p source has little detail left to lose. | Make it again at 720p. Upscaling the 480p file makes it bigger, not sharper. |
| Instagram muted my Reel | Instagram mutes copyrighted music it can't match to its library. | Download the Silent copy next to the video and post that, then add "Outstanding" from Instagram's music library if it is available in your region, lined up with the laugh close-up at 0:07; or post the version with sound to TikTok, where it usually stays. |
Tiny desk ai videos: Outstanding examples
Rust sweater lead
Full 19 s with the song · bald sax player in glasses · 480p
Navy polo and headset
Full 19 s with the song · band in headsets · 480p
Bleached crop and silver chain
Full 19 s with the song · five shots · 480p
The photo checklist
- One person per photo
- Facing the camera, eyes open
- Smiling, ideally with teeth
- Head and shoulders, not a full-body shot
- Even light, no sunglasses or heavy filters
- At least 300 px on each edge (a cropped screenshot from a fantasy app works if the face is sharp)
Only have screenshots of the players?
Crop each screenshot so one face fills most of the frame, then upload that crop. Avatars and team logos don't work: the model needs a real, visible face. Use photos of people who agreed to be in the joke.
More Tiny Desk AI fixes
Why does the singer's face keep changing halfway through my Tiny Desk AI video?
Usually the photo, and on 15-second models the 0:13 split. Use a sharp front-facing photo, or choose Seedance 2.5 to make the 19 seconds in one pass.
Our five-person video keeps putting the sax player's face on the keyboard player. How do I fix it?
Give every seat its own photo of one person: band mode has five seats (lead, keys, left and right backup, sax). If two faces keep swapping, fill fewer seats; an empty seat is played by an AI performer, which leaves the model fewer faces to keep apart.
Should I regenerate at 720p or upscale my 480p video?
Regenerate at 720p. Upscaling cannot recover detail the 480p video never had, and TikTok compresses it again on upload.
My video failed. Did I lose my credits?
No. A failed video refunds its credits automatically. If a photo fails the model's safety review, try a different, clearer photo.
Can I re-time the song without changing the video?
You rarely need to: the song is laid in sync with the five cuts. If you add the library track in Instagram instead, use the Silent copy and line the song up with the laugh close-up at 0:07.
Try again with a better photo
Pick a smiling, front-facing photo and the 720p setting.
Open the generator