What is the Love Is Gonna Save Us dance AI trend?
It is a run of AI videos in which people put themselves on the same party platform and do the same 16 seconds of electro arm moves to “Love Is Gonna Save Us”. The scene is always the same: a raised platform beside the DJ decks at an open-air night party, strings of warm lights overhead, the crowd below cheering and holding up phones. The dancer in the original clip wears an orange-red T-shirt and loose white trousers.
Three moves make it recognisable: arms crossed at the chest, one forearm swept across the face, then both arms flung wide. In the AI versions everything stays except the dancer, who becomes the person from a photo.
- The song. “Love Is Gonna Save Us” is by Benny Benassi presents The Biz, from the 2003 album Hypnotica. Most people now hear the remake by ARTBAT & Benny Benassi, whose official video is on ARTBAT's YouTube channel; a radio edit is on Spotify.
- The “dance guy”. When people search for the “Love Is Gonna Save Us dance guy” or the “Russian guy dancing”, they usually mean Cheba Man (@cheba_man), whose Tecktonik-style electro routines to the track spread across Instagram and TikTok in May 2026.
- Real or AI? Both. The dance is real and people learn it at parties; the trend you keep seeing is the AI part, where a photo replaces the dancer in one specific party clip. That clip is the built-in reference in the panel above.
Do you need a prompt? Template vs text, image and reference prompts
This is where most attempts go wrong. The trend is one clip, so what matters is whether your tool can see that clip.
| Way to make it | What you give the tool | Same stage and moves? | The track |
|---|---|---|---|
| Text-to-video prompt | A written description | No: the model invents a stage, a crowd and its own dance | Add it yourself in an editor |
| Image-to-video (photo + prompt) | Your photo and a description | No: your face, but improvised moves and a made-up party | Add it yourself |
| Reference video + photo + prompt | The party clip, your photo and a prompt like the one below | Close, if the tool has motion control or a reference mode | Usually silent; add it after |
| This page's template | One photo | Yes: the clip and the instructions are built in | Kept from the clip |
In short, a prompt alone can't recreate the trend; a reference video can. If you already have the clip and a reference-capable tool, use the prompt below. If not, upload a photo above and skip the prompt entirely.
A copy-ready Love Is Gonna Save Us dance AI prompt
For tools that accept a reference video plus one image, such as motion-control or reference modes. Attach the party clip as the video and your photo as the image, then paste this. Rename “reference video” and “reference image” to whatever your tool calls its inputs (for example @Video1 and @Image1).
Use the reference video for motion, timing and camera only. Put the person from the reference image on the raised party platform in place of the dancer, head to toe, in every frame. Keep that person's own face, hair, skin tone, build and clothes from the image. If the image stops at the shoulders, continue the same outfit down to the shoes. Copy the dance beat for beat: arms open, then a clap at the chest; both arms straight up; fast wrist rolls and forearm crosses in front of the face; arms flung wide with pointing fingers; swings across the body; one arm whipped overhead; a single high knee; quick sweeps past the face, ending with open arms. Keep the night sky, the string lights, the DJ decks and the crowd below filming on their phones. One low, slightly shaky handheld shot, no cuts, normal speed. The DJ and the crowd are generic strangers. One dancer only, natural hands, no text, logos or watermark.
Works best with the full-body photo described below. Keep the clip at 16 seconds or less; many reference modes stop at 15.
No reference video? This text-only prompt gets the look of the scene, but the model will choreograph its own version of the moves:
Vertical phone video at an outdoor night party. A person stands on a raised platform next to the DJ decks under strings of warm lights; the crowd below cheers and films on their phones. To a driving electro beat, they do a sharp Tecktonik-style arm routine: arms crossed at the chest, a forearm swept across the face, then both arms flung wide. Low handheld angle looking up, one continuous shot.
The beat sheet (also a quick tutorial)
Where each move lands in the 16-second clip. Use it to check a result, to time a prompt for a shorter cut, or to learn the routine yourself. Shorter videos are the opening of the dance.
| Time | Move |
|---|---|
| 0–1 s | Arms out to the sides, then the hands snap together at the chest |
| 1–2.5 s | Both arms shoot straight up; one stays raised |
| 2.5–5 s | The signature part: wrist rolls in front of the face, forearms crossing at the chest, a forearm swept across the face |
| 5–7 s | Both arms flung wide, fingers pointing out over the crowd |
| 7–11 s | One arm up by the head, then swings across the body and back to the chest while the crowd raises their arms |
| 11–12.5 s | An arm whipped up over the head and out, then one high knee lift |
| 12.5–16 s | Fast sweeps past the face and overhead, finishing with the arms open (16-second version only) |
Single portrait or full-body photo?
One photo is all it takes, and you don't need a video of yourself. What the photo shows decides how much the model has to guess:
| Your photo | What you get |
|---|---|
| Full body, facing the camera, one person | The closest match: your face, clothes and shoes on the platform. Best choice. |
| Waist-up or half-body | Works; trousers and shoes are invented to match your top |
| Face-only selfie | Works for the face, but the whole outfit is made up and the face is more likely to drift on the fast arm sweeps |
| Group photo | The model may pick the wrong person: crop it to one |
| Sunglasses, cap or heavy filter | Identity drifts and the result looks less like you |
How much does a Love Is Gonna Save Us dance AI video cost?
You pay per video in credits, and the price is on the Generate button before you start. It is one short clip from a fixed template, not a music-video production: the cheapest is 8 credits (5 seconds on Seedance 2.0 Mini at 480p) and the most expensive is 495 (the full 16 seconds on Seedance 2.5 at 1080p).
| Model | 5s | 10s | 15s | 16s |
|---|---|---|---|---|
| Seedance 2.0 Mini · 480p | 8 | 15 | 23 | — |
| Seedance 2.0 Mini · 720p | 16 | 32 | 48 | — |
| Seedance 2.0 Fast · 480p | 16 | 32 | 48 | — |
| Seedance 2.0 Fast · 720p | 35 | 69 | 104 | — |
| Seedance 2.0 · 480p | 27 | 54 | 81 | — |
| Seedance 2.0 · 720p | 58 | 116 | 174 | — |
| Seedance 2.0 · 1080p | 146 | 291 | 436 | — |
| Seedance 2.5 · 480p | 39 | 78 | 117 | 124 |
| Seedance 2.5 · 720p | 88 | 175 | 262 | 279 |
| Seedance 2.5 · 1080p | 155 | 310 | 464 | 495 |
| MiniMax H3 · 720p | 39 | 77 | 116 | — |
“—” means the model doesn't offer that option: the full 16 seconds is Seedance 2.5 only, and MiniMax H3 renders 720p here. On the Pro monthly plan (780 credits for $29), one credit costs about $0.037, so the full dance at 480p on Seedance 2.5 uses about $4.61 of credits, and $10.37 at 720p.
| Free | Signing up, uploading a photo and playing the reference clip. New accounts get 3 free credits, which is not enough for a video: the cheapest one costs 8. There is no unlimited free version. |
|---|---|
| Smallest purchase | $5 Trial Pack = 24 credits: one 5–15-second Seedance 2.0 Mini video at 480p (8–23 credits), not the full 16 seconds (124 credits) |
| Watermark | Videos made before your first purchase carry a small moving genjutsu-ai.app watermark; any purchase unlocks the clean originals, including videos you already made. |
| Generation time | Usually a few minutes. Seedance 2.5 reviews the photo and the clip first. You can leave the page; the video stays in your history. |
| Failed or rejected | Credits are refunded automatically |
When the result looks fake: what to change
| What you see | Likely cause | What to change |
|---|---|---|
| The face warps during the arm sweeps | Arms pass in front of the face; the photo is small or angled | A sharp, front-facing photo; 720p or 1080p; Seedance 2.5 or Seedance 2.0 |
| You look like a different person halfway through | Face-only or half-body photo, or a face hidden by glasses or a cap | A full-body photo with the face clearly visible |
| The moves don't match the original | A text prompt without the clip, or a cut that stops before the finale | Use the built-in clip; choose the full 16 seconds on Seedance 2.5 to get the knee lift and the ending |
| Blurry on Reels or Shorts | 480p, or a second export at a low bitrate | Pick 720p or 1080p and post the downloaded MP4 as it is, without re-exporting it |
| The crowd or DJ looks like someone real | They come from the original clip | Nothing: here they are always redrawn as invented people. In other tools, add that line to your prompt |
Love Is Gonna Save Us dance AI examples
The same routine on different people, on the rooftop platform and on the darker red-lit stage the dance also circulates on.
Friends' photos, the song and your data
- Other people's faces. Use your own photo or one of an adult who agreed to be in the video. No celebrities, strangers or minors; you confirm this before each generation.
- The music. Your video keeps the track from the reference clip, but buying credits is not a music license. For a personal post, TikTok and Instagram usually handle the sound through their libraries; for a brand ad, license the track or swap in a commercial sound.
- Your photo. It stays a private input on your account and is never shown on the site. See the privacy policy for retention and deletion.
Genjutsu AI is an independent tool, not affiliated with Benny Benassi, ARTBAT, Cheba Man, the dancer in the reference clip, Higgsfield or CapCut.
More one-photo and two-photo trend templates: the Bulin 47 street freestyle, the Toc Toc dance, the two-person Alors on Danse studio meme and the Raindance pier duet. For writing your own instructions on any clip, see the Genjutsu AI prompt guide, or start from the full editor with the Hotel Lobby clip.
Last updated: 2026-10-12