
A lanky man in a black coat walks through Manhattan with a coffee in one hand and a closed umbrella in the other. He hooks a manhole lid shut with his umbrella before a woman on her phone walks into the hole, fishes a coin out of a fountain for a little girl, and never once changes his expression. Sting's "Englishman in New York" plays underneath. That 47-second clip passed 36.8 million views on X in under three weeks, and the reposts on TikTok kept going after that.
This guide shows how to make an AI music video in the same hand-drawn style with your own character and an original song.
Key Takeaways
- The viral umbrella-man animations areantrofrog's original series, not a CapCut template anyone can tap and reuse.
- antrofrog's credits for No Ordinary Love list Midjourney 8.2 for the main character and Seedance 2.5 for the animation.
- Design one original 2D character first, then animate each short gag as its own beat.
- Sting and Sade recordings are copyrighted, so generate an original song before you cut a single frame.
- Every shot of this MV came out of the Atlas Cloud video generator running Seedance 2.5.
What Is the AI Animated Music Video Trend Behind "Englishman in New York"?
The viral clips in this trend trace back to one creator. antrofrog, whose bio reads "an artist with zero skills, so I just use AI," joined X in September 2026.
Four of his AI animated music video shorts carry the trend, and all four star the same character, George. Each one sets him loose in a city and cuts his quiet good deeds, or his quiet courtship, to a well-known song.
| Film | Song | Posted | Length | What happens |
|---|---|---|---|---|
| Smooth Operator | Sade | X, September 26, 2026 | 44 s | George fixes small problems around London: a cat in a tree, a falling flowerpot, a stolen handbag |
| Smooth Criminal | Michael Jackson | TikTok, October 26, 2026 | 44 s | Two figures in yellow raincoats pick a fight with him on a wet street at night and lose |
| No Ordinary Love | Sade | X, October 28, 2026 | 59 s | George courts a woman in a red dress across rainy London |
| Englishman in New York | Sting | X, October 30, 2026 | 47 s | George helps strangers in Manhattan with an umbrella and a coffee |
Lengths come from the copies on his own accounts. As of November 16, 2026, the No Ordinary Love post on X showed 22.3 million views and Englishman in New York showed 36.8 million.

Stills: Smooth Operator, Smooth Criminal, No Ordinary Love and Englishman in New York by antrofrog.
The TikTok wave is mostly reposts. The most-saved one we found is a 118-second compilation that stitches all four films together under the #noordinarylove tag. By November 16 it had 11.9 million plays and more than 155,000 saves.
It doesn't link a CapCut template or a TikTok effect, and neither do the other large reposts we checked. To get a video like this, you have to make one.
Not everyone loved it. Creative Bloq covered the backlash to Smooth Operator on September 28, 2026. The piece called the mix of art styles disjointed, and commenters it quoted accused the character designs of borrowing from other films. That's a good reason to design your own character instead of cloning George.
How to Make an AI Music Video on Atlas Cloud
antrofrog listed his tools in a reply under No Ordinary Love: Midjourney 8.2 for George's design, Krea 3 for the woman in red, Seedance 2.5 for the animation. Atlas Cloud puts Midjourney V8.2, Seedance 2.5 and Suno behind one login and one balance, so the workflow below runs without switching sites.
Our test MV follows Inês, a tall postal carrier with a red bicycle, through one morning in Lisbon. It runs about 45 seconds over an original 1980s-style pop song.
Step 1: Design an Original 2D Character in Midjourney V8.2
Open the Midjourney V8.2 playground and paste the prompt below. Set the aspect ratio to 16:9 so the still matches your video frame. Each run returns four options.
plaintext1A tall slender young woman postal carrier stands beside an old red bicycle on a steep sunlit Lisbon street, full body, head about one eighth of her height, sharp black bob with a straight fringe, calm half-smile, forest-green belted wool coat over a cream turtleneck, brown leather mail satchel across her body, yellow tram and blue azulejo tiles behind her, flat 2D hand-painted animation still, clean ink linework, gouache texture, muted warm palette with red accents, soft morning side light, wide cinematic composition

Choose the frame with a clear face, clean hands and every prop in view. We kept option 4: she faces the camera, and the bicycle bell that one of the later gags needs is right there on the handlebar. Every later shot points back to this still, so rerun now if the coat, haircut or satchel is off.
Method 1: Animate Each Beat in the Seedance 2.5 AI Music Video Generator
Seedance 2.5 accepts an audio file as a reference. Feed it a slice of your song and it times the motion to that track's beat. You write the gags, and the model handles the timing.
Open the Seedance 2.5 video generator. Seedance 2.5 is already the selected model on the Video Generation tab.
- Under Upload other references, add your character still and a 15-second slice of the song.
- Paste a three-shot prompt that cites them as @Image1 and @Audio1.
- Open Settings and choose 16:9 and 1080p.
- Set Duration to 15s.
- Switch Audio off, because the finished song goes on in the edit.
- Click Run.

Here is the prompt for the first 15 seconds. The other two runs use the same opening lines with new shots.
plaintext1@Image1 is the postal carrier. Keep her face, black bob, olive-green coat, leather satchel and red bicycle exactly the same in every shot. Flat 2D hand-painted animation, clean ink lines, gouache texture, no lip movement, no dialogue, no text. Time every cut and movement to the beat of @Audio1. 2Shot 1 (0-5s): wide shot, she coasts down a steep Lisbon hill on the red bicycle as a yellow tram passes and rings its bell, letters fluttering in her front basket. 3Shot 2 (5-10s): a laundry line snaps above a narrow alley; a white shirt floats down onto her shoulder and she pegs it back on the line without stopping. 4Shot 3 (10-15s): two tourists turn a huge paper map upside down; she folds it into a paper airplane and throws it, and it glides straight toward the right street.
fbX1vV3xv4U
Three runs of 15 seconds give you a 45-second MV. If one gag misfires, rerun only that 15-second block.
Method 2: Chain Midjourney and Seedance 2.5 Through the API
The same pipeline runs from a script, which helps when you want ten takes of each block.
Step 1: Create and copy an API key in the Settings page of the Atlas Cloud console.

Step 2: Check the request format in the Atlas Cloud API docs. Seedance 2.5 reference-to-video takes reference_images and reference_audios as lists, and the prompt cites them as @Image1 and @Audio1 in list order. One request takes at most 30 seconds of reference audio.
Step 3: Submit the clip, then poll for the file. The prompt is shortened here. Paste the full one from Method 1.
plaintext1curl -X POST https://api.atlascloud.ai/api/v1/model/generateVideo \ 2 -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \ 3 -H "Content-Type: application/json" \ 4 -d '{ 5 "model": "bytedance/seedance-2.5/reference-to-video", 6 "prompt": "@Image1 is the postal carrier. Keep her face, black bob, olive-green coat, leather satchel and red bicycle exactly the same in every shot. Flat 2D hand-painted animation, no lip movement, no dialogue, no text. Time every cut to the beat of @Audio1. Shot 1 (0-5s): ... Shot 2 (5-10s): ... Shot 3 (10-15s): ...", 7 "reference_images": ["https://your-cdn.com/postal-carrier.png"], 8 "reference_audios": ["https://your-cdn.com/song-part-a.mp3"], 9 "duration": 15, 10 "resolution": "1080p", 11 "ratio": "16:9", 12 "generate_audio": false 13 }'
plaintext1curl https://api.atlascloud.ai/api/v1/model/prediction/<prediction_id> \ 2 -H "Authorization: Bearer $ATLASCLOUD_API_KEY"
Poll until the status reads completed, then download the video from outputs. To generate the character from the same script, send the Step 1 prompt to midjourney/v8.2/text-to-image through the generateImage endpoint with the same key. One key covers every model, so switching models mostly means changing the model string.
Write an Original Song for Your AI Music Video
Here's the thing about the originals: the songs were the hook, and you can't borrow them. "Englishman in New York" and "No Ordinary Love" are copyrighted compositions and copyrighted recordings. Make the song first, then build the pictures around it.
Suno chirp-v6 on Atlas Cloud has a custom mode that takes your lyrics, a style line and a target length. Open its model page and work down the form. Describe the era and instruments, never an artist's name.
- Paste your lyrics into Prompt.
- Switch on Custom Mode, so the prompt is sung as lyrics instead of read as a description.
- Set Variety to normal, which keeps the two takes distinct without pushing them far from the prompt.
- Pick the vocal gender.
- Give the song a title.
- Write the style line.
- Set Duration a little longer than the MV, so you have room to trim.
- Open Additional Settings and list what to keep out under Exclude Styles.
- Click Run.
plaintext1[Verse] 2Morning on the hill, the tram bell rings 3I carry other people's everythings 4A hat, a cat, a letter gone astray 5I hand it back and pedal on my way 6[Chorus] 7Oh, Lisbon in the light 8Every small thing put right 9No one needs to know my name 10Tomorrow I'll ride by again
| Field | What we used |
|---|---|
| Variety | normal |
| Vocal Gender | Female |
| Title | Ride By Again |
| Style | 1980s sophisti-pop, smooth saxophone, fretless bass, soft drum machine, warm female vocal, mellow, nostalgic, 96 bpm |
| Duration | 50 seconds |
| Exclude Styles | rap, EDM, distorted guitar, autotune |

Each run returns two takes, so check which one gets to the words fastest. Our first take spent 16 seconds on an intro before the first line. The second started singing at 1.5 seconds and reached the chorus at 28.7, so we kept it.
By our measurement it came out at about 99 BPM, close to the 96 we asked for, which put the 15- and 30-second marks within half a second of a bar line. We trimmed it to 45 seconds plus a short fade and cut three 15-second slices at those marks.
2D AI Animation Prompts That Keep the Hand-Drawn Look
Learning how to make AI animation that reads as hand-drawn mostly means saying the same few words every time. These are the lines that held the look across our three runs.
| Prompt line | Why it's there |
|---|---|
| Flat 2D hand-painted animation, clean ink lines, gouache texture | Names the same look in every run |
| Keep her face, black bob, olive-green coat, leather satchel and red bicycle exactly the same | Lists every trait you'd notice if it changed |
| No lip movement, no dialogue, no text | Asks for closed mouths and no lettering |
| One gag per five-second shot | A single cause and effect fits in five seconds |
| Time every cut and movement to the beat of @Audio1 | Lets the reference track set the rhythm |
Here are the six shots for the other two runs. Each block follows its slice of the lyrics: the second slice sings "a hat, a cat, a letter gone astray," so those three gags land in Run B.
Run B, 15 to 30 seconds
plaintext1Shot 1 (0-5s): a gust lifts an old man's straw hat; without braking she catches it on a rolled newspaper and drops it back on his head, and he blinks. 2Shot 2 (5-10s): an orange cat is stuck on a tiled window ledge; she rings her bicycle bell twice and the cat leaps neatly into her front basket. 3Shot 3 (10-15s): she sets the cat down at the feet of a small girl in a yellow dress, hands a stray envelope to a woman in a doorway, and pedals on.
Run C, 30 to 45 seconds
plaintext1Shot 1 (0-5s): a baker's tray of custard tarts tips off a café counter; she slides past and catches every tart on her satchel flap, then sets the tray back. 2Shot 2 (5-10s): sunset at a hilltop viewpoint over red roofs and the river; she finds one envelope addressed to her, opens it, and smiles. 3Shot 3 (10-15s): she rides away downhill into the golden light while a trail of letters lifts from the basket and flies after her like birds.
If you want an anime look instead of a storybook one, our Seedance 2D animation guide covers style anchors for that.
From Clips to a Finished AI Cartoon Music Video
The edit is where an AI cartoon music video starts to feel directed. Your three clips were timed to three slices of one song, so this step is mostly lining them up.

- Drop the trimmed song on the timeline first, with a short fade at the end.
- Place clips A, B and C at 0, 15 and 30 seconds.
- Trim a few frames where a gag lands early or late. Nudge the video, never the music.
- Mute the clips' own audio if any came through, and export at 16:9. Make a 9:16 crop too if you're posting to TikTok.
WPRbcUpfzAs
If you'd rather make a music video with a real singer and lip sync, the Seedream and Seedance MV workflow walks through that version.
Frequently Asked Questions
Can I use Sting or Sade songs in my AI music video?
Not without a license. A famous song is two copyrighted works at once: the composition and the recording. Posting one under your animation can get the video muted or removed. If you want that same mood without the legal risk, generate an original track and describe the era and instruments in your prompt, never the artist.
Is there a CapCut template for the Englishman in New York AI animation?
No. The big TikTok reposts of Englishman in New York and No Ordinary Love that we checked link no CapCut template and no TikTok effect. They are antrofrog's original films, reposted or stitched together. You have to build your own version from a character still and short animated shots.
What AI did antrofrog use for No Ordinary Love?
His credits reply on X lists Midjourney 8.2 for George's character design, Krea 3 for the woman in the red dress, and Seedance 2.5 for the animation. He hasn't published his prompts or shot list.
Can I make an AI music video for free?
Not with this workflow. Each step on Atlas Cloud bills per run, and Seedance bills by the second of video. The next answer adds up a full 45-second MV.
How much does a 45-second AI music video cost on Atlas Cloud?
As of November 16, 2026, one Midjourney V8.2 image run costs $0.086 for four options, and one Suno chirp-v6 run costs $0.132 for two takes. A 15-second Seedance 2.5 reference-to-video clip at 1080p costs $8.90, which is the price after a 20% discount shown on the model page with no end date listed (list price $11.12).
Our three clips plus one image run and one song run came to about $26.90 before rerolls. Rendering the same clips at 720p brings the Seedance part down to $13.56.
Conclusion
antrofrog's umbrella man worked because the gags landed on the music and the character never changed. You can get both of those things with your own character and your own song. Lock the character in one still, hand Seedance a slice of the track for each 15-second block, and keep the prompt lines that hold the hand-drawn look.
Once you've done one, the second goes much faster. The character still and the prompt header stay the same, so the next song only needs new gags and a new beat. That's how to make an AI music video that's yours from the first note to the last frame.






