What You're Trying to Make
One photo in, and out comes you — or your baby, or your dog — standing up in your own room, belting a Neapolitan hook straight to camera with the exact hand choreography from the viral clip: the finger point, the kiss blown on the beat, the confident sway. The audio is the real hook, and every gesture lands on it perfectly, because the performance isn't described to an AI — it's transferred.
The song is "NUN CAGNAM MAJE" by Christian Liguori & Rivreck — Neapolitan for roughly "we never change" — released July 10, 2026 on Epic. The clip that carried it was an AI-generated video of Erling Haaland singing it in a bathroom, pointing and blowing kisses in time with the "NUN CAG" hook. It tore through South American and global football TikTok during the July 2026 World Cup, and the participation format followed immediately: put yourself in, doing the same moves.
Fastest way — Nun Cagnam Maje on Starrd does it in one tap: upload a photo and it renders the full 14-second performance — your subject, your photo's own background, the real hook, every gesture on the beat. Want to know how it works first? Read on. ↓
Is This a Trend? (Yes — It's the Song of Football TikTok Right Now)
Very much so. The AI bathroom clip took off on the back of Haaland's standout World Cup run — two goals against Brazil in the round of 16, Norway's quarter-final on deck — and the "NUN CAG" hook became the sound of football TikTok from Buenos Aires to Naples. CapCut is already shipping one-click templates on the audio, which is the reliable tell that a sound has crossed from meme to format.
Worth knowing there are two AI-Haaland memes running at once and they get conflated: this one (the bathroom singing clip) and the Techno Viking face-swap — Haaland's face on the 2000 Fuckparade dancer — that crested around July 8. Different meme, same "AI Haaland" moment — this is the one with the song.
Here's the gap the self-insert version fills: reposting the original is just reposting. Putting your own face — or your baby's, or your dog's — on the exact performance is participation, and the versions that travel are the ones where the subject swap is the joke. The window on trends like this is weeks, not months, so the fastest path matters.
The Fastest Way — Use the Nun Cagnam Maje Template on Starrd
The Nun Cagnam Maje template packages the whole thing into a single upload:
- Pick one clear photo. You, your kid, your pet — full-body or waist-up, decent lighting, one subject.
- Open the template in the Starrd app or web library.
- Upload and tap generate. Starrd stands your subject up in the photo's own setting, then maps the exact viral performance onto them — 14 seconds, vertical 9:16, real hook audio included.
Nun Cagnam Maje
Drop any photo and watch them belt the viral Neapolitan anthem straight to camera — the pointing, the kiss on the beat, all mapped from the original performance and synced perfectly. Works on you, your baby, or your pet. No prompt to write.
The rest of this guide is for people who want to build it by hand.
Or, Build It Yourself — This Is a Motion-Control Format
The single most important thing to understand: you cannot text-prompt your way to this video. A prompt like "person sings and gestures to a Neapolitan song" produces generic swaying that lands nowhere near the beat. Every version you've seen that syncs is built with motion control (motion transfer): a fixed driving clip supplies the exact choreography, and a still image supplies the character. Same recipe as the kung fu cat trend.
You need three things:
- The driving clip — the original performance, trimmed to the hook.
- A motion-control model — Kling 2.6 Motion Control is the reliable pick; it maps the clip's motion onto your still and carries the clip's audio through to the output.
- A clean upright still of your subject — this is where most attempts die. Motion control estimates a skeleton from your image, so the subject must be fully upright, full body, one clear unobstructed silhouette. A seated, cropped, or overlapping subject fails outright. Pets must be stood up on two legs first.
STILL PREP (run on an image model with your photo as reference): Transform the subject from the reference photo into a full-body shot of the SAME subject standing tall and fully upright — an animal stands on both hind legs like a person — not sitting or crouching. Arms raised to chest height and separated, hands free and unobstructed, as if about to sing and gesture passionately. Preserve the exact identity (face, hair or fur, skin or markings, clothing) from the reference — only the pose changes. One clear unobstructed silhouette. Keep the exact same background, setting, and lighting as the original photo. Full body head-to-feet, vertical 9:16, photorealistic. No text, no logos, no watermarks.MOTION PASS (motion-control model, driving clip + the still above): The subject sings passionately straight to camera, lip-syncing with heartfelt emotion, gesturing expressively with both hands like a soulful ballad singer. Confident, dramatic, theatrical energy. Smooth natural motion, stable consistent character.
Note how short the motion-pass prompt is. With motion control, the timing all comes from the driving clip — the prompt only sets energy and character stability. Long beat-by-beat prompts do nothing here; that effort belongs in the still, whose framing and silhouette decide whether the skeleton maps cleanly.
Common Mistakes That Tank Your Video
- A seated or cropped reference. Motion control needs a full upright body to build a skeleton. Waist-up photos work as input only if you first generate the full-body standing still from them.
- Text-prompting the performance. You'll get generic swaying that misses every beat. Use motion transfer.
- An obstructed silhouette. A hand in front, a second person, a held phone — anything overlapping the subject breaks the pose estimate. One clean silhouette.
- Swapping the background. The charm of your version is your own setting — a studio backdrop makes it look like a copy of the original instead of your take on it.
- Claiming the original is real footage. It isn't — the bathroom clip is AI, the whole trend is openly synthetic, and your version should carry the platform AI label too.
Frequently Asked Questions
What is the trend? A Neapolitan song hook plus an AI clip of Erling Haaland singing it in a bathroom with pointing and kiss-blowing gestures — and a participation format where one photo puts you (or your baby, or your pet) into the same performance.
What does the title mean? "Nun cagnam maje" is Neapolitan for roughly "we never change."
Why does every version sync perfectly? Motion control — a fixed driving clip supplies the exact choreography, so the beat-sync is identical in every version. Text prompts can't do that.
Can I make my dog or baby do it? Yes, and those versions usually outperform the self-insert. One photo either way.
Do I need to label it as AI? Yes — all major platforms require it, and this trend is openly AI anyway.
Related Reading
- How to Make a Kung Fu Cat AI Video — the other motion-control comedy format: your pet's karate meltdown, beat-for-beat.
- How to Make an AI Dance Video — the full motion-transfer dance library, same one-photo workflow.
- How to Make the World Cup AI Video — the other big July 2026 football format: run onto the pitch and score.
- Viral AI Video Trends (2026): The Monthly Roundup — every trend worth making this month.
