The Somebody That I Used To Know AI trend is a 15-second duet where two people, painted into a pastel triangle wall like Gotye and Kimbra in the 2011 music video, sing the song at each other. To make it with your own faces, upload two photos on videosi, one of you and one of your partner or friend, and download a finished MP4 that already carries the original song.
This guide goes deeper than the page FAQ: where the format comes from, how the original video was shot, what the lyrics mean, how the AI version breaks into shots, how to choose photos, and what to do about lips, rights and cost.
Key takeaways
- The trend copies the painted-wall look of the 2011 Gotye and Kimbra video; the earliest AI duet we found is dated 3 October 2026.
- You need two photos of adults, one per singer, or one photo of both; no prompt is needed.
- The video has the original song, but the mouths are not synced to the lyrics.
- A standard 15-second video is 13 credits, and packs start at $6.90.
- Post it with the AI label on and use the song only where the platform allows.
In this guide
- Where does the painted-wall duet come from?
- How did Gotye and Kimbra film the original painted wall?
- What do the lyrics mean, and why does the ex theme fit?
- How does the AI version break the duet into shots?
- How do I pick the two photos?
- Why do the mouths not match the lyrics, and what can I do?
- Is it safe to post, and should it carry an AI label?
- How is this different from a CapCut template?
- Can I make it with ChatGPT, Veo or Kling instead?
- What does the AI duet cost, and what is not accepted?
Where does the painted-wall duet come from?
The painted-wall duet is a 2026 TikTok format that borrows the look of the 2011 Gotye and Kimbra music video. The earliest AI version we found is a two-person clip dated 3 October 2026, and CapCut template tutorials spread widely from 3 November 2026. We could not confirm who started the wave, so we do not name a creator.
What we can say is how the format is built. Two people stand close, face to face, in front of a pastel wall of triangles. Their skin is covered in matching patches of colour so they seem to be part of the wall. They sing the song at each other. The Somebody That I Used To Know AI page on videosi covers the version with two of your own photos.
The song gives the format its reach. Somebody That I Used To Know by Gotye featuring Kimbra came out in 2011, and many viewers recognise the opening bars at once.
How did Gotye and Kimbra film the original painted wall?
The original video was directed, produced and edited by Natasha Pincus, with cinematography by Warwick Field, and its look is body painting plus stop motion. Wikipedia credits Emma Hack for the body painting and Howard Clark for the backdrop, which is based on a 1980s artwork by Gotye's father, Frank de Backer.
The effect works like this: the singers are painted in the same shapes and colours as the wall, and Gotye's skin is gradually painted into the background through stop-motion animation as the song goes on. The Australian Centre for the Moving Image also describes the animation as stop motion (ACMI).
What do the lyrics mean, and why does the ex theme fit?
The song is about the aftermath of a breakup, told from both partners' points of view. Gotye has said it draws on his own past relationships (Wikipedia). The first verse is his; the second answers from her side, which is why it is a duet.
That structure is what makes the trend work. The man sings about how things ended, the woman replies, and the two stare at each other while the wall swallows them. People add their own meaning on top: a real ex, a film couple, two friends pretending to be one. The ache of the song stays the same, and the comedy comes from who is in the pair.
How does the AI version break the duet into shots?
The AI duet follows six beats over 15 seconds, copied from the reference clip. First comes a wide two-shot against the triangle wall, one singer on the left and the other facing them in profile on the right. Then a medium two-shot where the first singer sings with eyes closed while the other listens and answers.
Around the middle the camera cuts to a close-up of the second singer's profile, face painted in rose and sage patches, singing toward the first. After a closer shot of the first singer, the camera pulls back to a wide two-shot again, the two small against the wall. The last seconds are a medium two-shot and a final close-up as the song settles.
Two details matter when you plan photos. Photo 1 is the person on the left and gets the long medium shots and the final close-up. Photo 2 is on the right and gets the profile close-up. The camera stays almost still, so faces carry everything. Open the Somebody That I Used To Know AI page to see the example clips beat by beat.
How do I pick the two photos?
Use two clear, front-facing, evenly lit photos of adults, one face per photo, with hair and shoulders in frame. The clip has close-ups of both faces, so sharpness matters more than anything else. A photo cropped tight to the face makes the model invent hair and build.
- Keep each photo to one person. If a photo shows two people, both are used, and the clip only has two singers.
- Pick photos where the two people look clearly different in hair or colouring, so the faces do not blend.
- Avoid sunglasses, hats and heavy filters, which can shift between shots.
- Put the person you want on the left in slot 1.
You can also switch to one photo of you both, with you on the left. If a face drifts in the result, replace that photo with a sharper one and run it again.
Why do the mouths not match the lyrics, and what can I do?
The finished video carries the original song, but the mouths follow the movement of the reference clip and are not synced to the lyrics. That is how the current version works. The people sing, the faces move in time with the shots, and the audio is the real track, but a viewer who watches the lips closely will notice the mismatch.
The practical fixes are simple. Post it as a lip-sync style clip, where a loose match is part of the joke. If you need exact sync, a dedicated lip-sync editor is a separate step that videosi does not do.
Is it safe to post, and should it carry an AI label?
Post it with the platform's AI-generated content label switched on, and use the song only where that platform's music rules allow it. The song is copyrighted by its owners, and videosi does not grant any licence to it. The MP4 carries the original sound because that is what the trend is built on, so check the music library of TikTok, Instagram or YouTube before you post. If a platform mutes the track, you can add the song from its own library.
Two more points. First, videosi does not accept photos of celebrities, public figures or minors, so the pair must be adults you know and who agree to appear. Second, the people in the video are real faces made to sing, so do not use anyone's photo without permission. This is general guidance, not legal advice.
How is this different from a CapCut template?
A CapCut template drops your photos into a fixed edit, while videosi generates a new video in which the faces actually move and sing. CapCut template tutorials for the trend surged on TikTok from 3 November 2026, and templates change often, so check inside the app. videosi is independent and not affiliated with TikTok or CapCut.
The difference shows in the result. A template places your photos into a fixed edit that someone else built. The AI version is a 15-second clip in which each person is painted into the wall and sings with moving lips, head and eyes. The cost is time and credits: a render takes a few minutes and uses credits, while a template is quick. The Somebody That I Used To Know AI generator needs only the photos.
Can I make it with ChatGPT, Veo or Kling instead?
You can try, but it takes many attempts. A general chatbot or video model can produce a duet in front of a patterned wall, yet the shot order, the matching paint on the skin and the two likenesses rarely hold at once. Models such as Veo or Kling can work from a good prompt and a reference, and you would have to write that prompt and supply a motion reference yourself.
videosi uses ByteDance's Seedance video model (Seedance 2.0 Mini by default, with Fast, Pro and Ultra one click away), a written prompt and a 15-second motion reference. All of it is built into the page, so you choose only photos and a format: 9:16, 1:1 or 16:9. The output is a ready MP4 with the song on it.
What does the AI duet cost, and what is not accepted?
A standard 15-second video (Standard model, 720p) costs 13 credits, and the video is not free: there is no free generation and no free trial. Credit packs start at $6.90 for 10 credits and never expire. Better models and higher resolutions cost more, and a failed render returns its credits.
On the input side, videosi accepts two photos of adults you have permission to use. It does not accept photos of celebrities, public figures or minors. The result has no watermark. The page shows the cost before you press Generate.
Ready to try it? Open the Somebody That I Used To Know AI page, add two photos and download your duet.
