How to Make the Tiny Desk AI Video (Outstanding Trend) Without the Looks Mixing Up

The Tiny Desk AI trend took off in October 2026. It uses Charlie Wilson's NPR Tiny Desk Concert from June 2023, where his band sings "Outstanding" (The Gap Band, 1982) packed behind the famous desk. On October 18 a TikTok parody put five college football coaches who had just lost into that band, and every weekend since has brought a new version.

Swapping four people at once is much harder than swapping one. Our first four runs failed, and the usual problem in this kind of video is that looks bleed between people: a cap, a jacket or a hairline from the original clip, or from another reference photo, ends up on the wrong person, so nobody quite looks like their photo. Below is what we changed to stop that, the exact prompt, and what it cost.

Why AI face swaps in this trend stop looking like your photos

The model overfits to what it sees most. In a video-to-video run it sees the original people in every frame, and your reference photos only once. So it tends to blend the two: it keeps the original coach's white cap on your friend, mixes the red jacket from the clip with the outfit in the photo, or drifts the face halfway between the original person and yours.

With four people it gets worse. The model also mixes the references with each other, so the hairstyle from image 2 can show up on the person from image 3. The result is a video where everyone looks a bit off, which kills the joke, because the whole point is recognizing who's in the band.

The two things we changed

Fix 1: Pixelate the faces in the source clip

We pixelated every face in the trend clip before using it. With no original face to copy, the model has to take the face from the reference image instead of averaging the two. The bodies, the fist pumps, the camera and the room stay readable, so motion and timing still come through.

Fix 2: Tell the prompt the original clothes are only labels

To tell the model who is who, you have to describe the original people ("the person in the red jacket"). That description is exactly what makes the model keep the red jacket. So the prompt says outright that those clothes are only for identification, that each new person takes the full look from their own photo, and that no hats or accessories from the original stay. Since one coach wore a cap that kept coming back, we added "No one wears a hat."

Method 1: Make it yourself with video-to-video AI

1. Save a clip. Save one Tiny Desk AI version from TikTok as an MP4 and trim it to the part you want. Video-to-video models take up to about 15 seconds; ours was 11.5 seconds, 720×1280, starting just after the first second so the clip opens on the band.
2. Pixelate the faces in the clip with any video editor (a mosaic or blur on each face is enough).
3. Pick four photos. One person per photo, face toward the camera, waist up with the outfit visible. The outfit in the photo is what they will wear in the video.
4. Set up the Video tab: engine Wan 3.0 Prime, the pixelated clip as source, your four photos as images 1 to 4, ratio and length matching the source, 480P, audio on.
5. Paste the prompt above and change the four position descriptions to match your clip. Press Generate; the result appears in My Videos, and failed renders refund their credits automatically.

Our test runs: what worked and where it slips

We ran the same prompt five times on October 26, 2026. Four runs failed (three on Wan 3.0 Prime, one on Wan 3.0) and returned no video; their credits were refunded automatically. The fifth, on Wan 3.0 Prime, worked: all four characters kept their own hair and outfits, the cap from the original never came back, and the camera moves, fist pumps and the song line up with the trend clip.

Where it slips: the person standing far left is small and often half out of frame, so likeness is weakest there. Put the photo you care least about in that slot and your best, sharpest face in the middle (the keyboard) or the close-up slot.

Method 2: Use the ready-made Tiny Desk template (four photos)

We packaged the working run as a template: the pixelated clip, Wan 3.0 Prime, the prompt above and the settings (12s · 480p · vertical · with sound). You only add the photos.

1. Open the Tiny Desk AI template. The template is already loaded.
2. Upload four photos into the four slots: standing left, at the keyboard, standing right, and the close-up.
3. Press Generate. The video shows up in My Videos when it's ready, so you can close the tab. Download the MP4, no watermark.

FAQ