Quick answer
Starrd's She's Staring At Me template turns three photos into @gnomeboys' 'so so mad' alley clip, set to Ne-Yo's 'Mad'. Photo 1 is the singer: filmed from the ground, they walk out of the sun, crouch over a small blue cigarette pack, pick it up, stare into the lens and sing 'wondering what she's thinking', then dance it out alone. Photo 2 is the one staring, always on the left, cold and silent behind a long stare. Photo 3 is the dancer, always on the right, who pushes the starer, throws both arms out and mouths 'nobody's talking'. All three walk off down the alley in the last shot. Ten shots cut where the original cuts, with the singer's lines lip-synced to the song. Everyone keeps their own face, hair and outfit. Works with pets. 20 seconds, vertical 9:16, Seedance 2.5, 480p, 200 credits (about $10).
The same dance, any subject — tap one to make it with your photo.
One of you sulks over a pack of smokes and sings Ne-Yo's "Mad" straight into the lens. One of you stares the camera down. One of you dances it off. Three photos put you and two friends in @gnomeboys' "so so mad" alley clip. Here it is:
She's Staring At Me
Three photos. Sulk, stare and dance it off in a sunny graffiti alley to Ne-Yo's 'Mad'. 20 seconds.
What You Get
- Three parts. The singer with the pack, the one staring on the left, the dancer on the right.
- Ten shots, cut like the original. The low ground shot, the extreme close-up, the over-the-shoulder, the two-shots, the solos and the walk-off, each cut on the original's frame.
- Lip sync where it counts. The singer sings "wondering what she's thinking" to the camera; the dancer mouths along on "nobody's talking".
- Your own look. Your faces, your hair and the clothes from your photos, in the clip's sunny graffiti alley.
- People or pets. Animals stand and walk upright on their hind legs, accessories on.
- 20 seconds, vertical 9:16, with the song, ready for TikTok, Reels and Shorts.
What the Video Actually Is
The clip is from @gnomeboys on TikTok, posted October 12, 2026 and captioned "so so mad" (137K+ plays, 11K+ likes and 600+ saves so far), the same crew behind the 12.5-million-play "Closer" car park dance battle. It's set to Ne-Yo's "Mad" (2008). Three friends in a sunny back alley of red-brick walls and graffiti: one in a light beanie crouches over a small blue pack and sulks, the other two (one in sunglasses, one in a flat cap and a tank top) stare back, then everyone dances it off and walks away.
Why it suits AI: it's a three-person bit with ten shots, a lip-synced line and a location you'd need to find. Most friend groups won't film that. Three photos cover all of it.
The 20 Seconds, Shot by Shot
| Time | Shot | What happens |
|---|---|---|
| 0:00 | Low angle from the ground | The singer walks out of the sun, crouches over a blue pack, picks it up and stares at it |
| 5.7s | Extreme close-up | The singer turns to stare into the lens, deadpan |
| 6.9s | Over the singer's shoulder | The singer holds out the pack; the other two stare back by the blue bins |
| 7.9s | Close two-shot | The one staring and the dancer look straight into the lens, unimpressed |
| 8.9s | Medium | The singer stands up with the pack and sings "wondering what she's thinking" |
| 10.8s | Wide | The dancer pushes the starer's shoulder, then shuffles a loose step |
| 13.1s | Medium | The singer dances alone, elbows up, fists by the head |
| 14.5s | Wide | The dancer throws both arms out and sways, mouthing "nobody's talking" |
| 16.3s | Close | The one staring stares on; the dancer's arm sweeps across the frame |
| 18.0s | Very wide | All three walk off down the alley, the sun behind them |
Your Look, Their Moves
- Your own outfits. A fedora and braces, a dress, a hoodie: whatever is in each photo walks the alley. Nobody gets the original's beanies, logo T-shirts, tank top or cargo pants.
- Hats and glasses only if they're yours. The original crew's beanies and sunglasses stay out of it unless your photo has them.
- Seen head to toe. The wide shots show everyone full length, so a photo that stops at the waist gets the rest of the outfit filled in to match its style.
- Pets stay pets. A chimp, a dog or a cat stands upright on real paws and keeps its own fur, glasses or chains.
Put the most expressive face first: photo 1 gets the close-up, the sulking and the sung line. The one staring (photo 2) never talks, so a deadpan friend is perfect there.
The Fastest Way
She's Staring At Me
Three photos, ten shots: the sulk, the stare and the dance-off in a graffiti alley. 20 seconds.
Three photos, 200 credits (about $10), no prompt. It renders on Seedance 2.5 at 480p, vertical 9:16, 20 seconds, with the song on top.
Build It Yourself
1. Turn the clip into a depth map. Run the clip through a depth-estimation model (Depth Anything V2 works). The depth map keeps every pose, every cut and the camera, and drops the faces, the clothes and the colour, so nobody from the original shows up in your video. Smooth the beanies and the flat cap into plain head shapes first, or hat brims carry through.
2. Make one character sheet. A vertical sheet with the empty alley on top and a four-view turnaround (front, three-quarter, side, back) of each of your three subjects below it, generated from your photos.
A photorealistic VERTICAL 9:16 CHARACTER REFERENCE SHEET laid out as four horizontal bands separated by thin black lines, with no text. Top band: an empty sunny back alley of grey asphalt between red-brick walls covered in colourful graffiti, blue wheelie bins, a bright blue sky and hard low sun. Below it, one band per subject: a 360-degree turnaround of the subject from reference image 1, then reference image 2, then reference image 3, full figure from head to feet, front, three-quarter, side and back, standing relaxed with the mouth closed, each in exactly the outfit from their own photo, with the alley softly out of focus behind them.
3. Drive it with the depth map, with the song inside. Feed the sheet, your photos and the depth video (with the song as its audio) to a model that copies motion from a reference video (we recommend Seedance 2.5). Then write what it can't read from a depth map:
- Who is who, shot by shot: the singer in the solos, the starer always on the left, the dancer always on the right, and who walks where in the last shot.
- The cuts, with times: a timed shot list, so each cut lands on the original's frame.
- The mouths: when the singer and the dancer sing, and that the starer's lips stay shut.
Lay the song back on top at the end: Seedance can drop or change music on its own, so mux the track yourself.
Common Mistakes
Driving with the raw clip. The original crew's faces, beanies and tattoos come straight through. Use the depth map.
Leaving the walk-off unassigned. In the last shot all three are small figures with the sun behind them, and the model happily draws the same person twice. Say who walks on the left, in the middle and on the right, and that each appears once.
Calling the prop the wrong thing. The depth map only carries a small shape in the hand. Name it (a small blue cigarette pack), or it becomes a drink can.
A silent driver. With no song inside the depth video, the singer's mouth drifts off the words. Put the track in the driver as well as on top.
Letting the starer talk. Without an explicit line, every face in the two-shots starts mouthing the lyrics. Say the one staring never sings.
Related Reading
- How to Make the 'Closer' by Ne-Yo AI Dance Battle Video: @gnomeboys' other Ne-Yo clip, a car park dance battle for two
- How to Make the 'I Need A Girl' AI Dance Video: a lip-synced intro, then a two-friend dance on the beat
- How to Make the 'Iconic By Mistake' AI Dance Video: another three-photo trend, a trio routine in unison
- How to Make the Wipe Me Down AI Dance Video: a two-friend dance in one take
- Viral AI Video Trends 2026: every trend we've turned into a template this year
