Alors on Danse AI Video: two photos, one white studio, and the Stromae–Jamel sketch, explained

An Alors on Danse AI video is a two-person remake of a white-studio comedy sketch: one person sits at a small table playing a mini keyboard next to an open laptop, head down and focused, while a second person beside them claps on the beat, points at the keys, laughs, dances, and leans right into the lens [reported]. The whole thing is made from two photos — nothing is filmed. Photo 1 is always the keyboard player; photo 2 is always the hype partner [reported]. Swap the order and you swap the joke.

The scene is not a real studio session, even though it circulated for years as a fake “making of” [reported]. It is a scripted sketch from Jamel Debbouze’s comedy DVD Made in Jamel, released on 1 December 2010, in which Jamel — the French comedian — bursts in on Stromae and pretends to teach him how to write “Alors on Danse”, taking credit for the hit note by note while Stromae plays along [reported]. The full sketch runs about six minutes; the part people share is the last twenty or thirty seconds, where the hook lands and the two of them rush the camera [reported]. The song itself was Stromae’s breakout single, released in Belgium in September 2009 and across Europe in February 2010, reaching number one in Belgium, France, Germany, Italy and many other countries [reported]. “Alors on danse” is French for “so we dance” [reported].

Do not confuse it with the other Alors on Danse meme. In 2021 a slow, swaying group dance to the same song spread on TikTok, with one creator’s clip passing 168 million plays [reported]. The 2026 trend is the studio sketch, not the group sway — two people, one keyboard, one white room.

The AI remakes started in late September 2026 [reported]. The earliest AI take one guide could find was posted on 25 September, a parody starring an Indonesian politician [reported]. Within days a version putting OpenAI’s Sam Altman and Anthropic’s Dario Amodei in the two roles was everywhere, and Fast Company wrote about the Altman–Amodei AI-video universe on 7 October 2026, describing Altman at the keyboard with Amodei offering beat advice in a white room [reported]. A TikTok upload of the original sketch from 1 June 2026 had 10.1 million views by 10 October [reported]. The two AI reels the template sites copy most are @kennyslowbird’s two AI parrots, posted 3 October — 805,700 views and 64,600 likes by 10 October [reported] — and Italian comedy duo Marta e Gianluca’s version with their own faces, posted 5 October, with 101,800 views by 10 October [reported]. The hashtag #alorsondanseai starts the same day as the Marta e Gianluca post [reported].

There is no canonical length, and that is the first thing to know before paying. The sites this page read on 8 November 2026 sell different cuts of the same sketch: 10 seconds on one dedicated site [reported], 11 seconds for people and 15 for pets on another [reported], 12 seconds on a third [reported], 15-second or full 16-second cuts on a fourth [reported], 10, 15 or 20 seconds on a fifth [reported], and about 30 seconds — the full-sketch cut — on two more [reported]. One general editor lets you pick any window from 5 to 30 seconds [reported]. The modal template is 10–16 seconds; 18–30 seconds is the long end, not the typical video. The frame splits the same way: most templates are landscape 16:9 like the original sketch [reported], while one general editor is vertical 9:16 [reported] and one template site offers 9:16, 1:1 or 16:9 for the same 12-second clip [reported].

The beats below are the 12-second cut, the most precisely timed version any site publishes [reported]. The timings are approximate — the beats, not the seconds, are what to check when a generation breaks.

0–1 seconds — arm up. The keyboard player is already on the keys, head nodding; the hype partner throws one arm up over their head [reported].

1–4 seconds — producing. The partner leans over the keyboard, pointing at the keys and directing with quick flicks, while the player stays focused [reported].

4–5.5 seconds — the snap. The partner straightens up, snaps their fingers and bounces on the beat [reported].

5.5–9 seconds — the hand roll. A rolling hand in the air and a proud, knowing face while the player keeps playing [reported]. In the 15-second telling of the same scene this whole stretch is the “hype” — fists on the chest in time with the beat, pointing, a finger to the sky — and at 9 seconds a hard cut opens to the whole white studio under round ceiling lights [reported].

9–12 seconds — turn and grin. The partner swings toward the lens and fills the right of the frame, ending on a huge grinning close-up with the player small behind them giving a thumbs-up [reported]. In the 15-second cut the ending is instead a rush: one of them charges the lens with a dead-serious face and pumping fists while the other dances at the back [reported]. In the human 11-second reel the ending is different again — both jog side by side toward the camera while it pulls far back into the empty white studio [reported] — and the ~30-second versions keep the whole sketch ending, jog included [reported].

When a generation breaks, check the beats in this order: the partner’s close-up at the end (the face that fills the frame), then the hand roll, then whether the two faces stayed apart after the cut. The cut at 9 seconds is where faces most often drift back to the stand-in or swap sides [reported].

The joke is a casting contrast, not a dance. One person plays it straight at the keyboard; the other commits far too hard beside them [reported]. Every couple, friendship and workplace has a calm one and a loud one, so the casting writes itself — and the caption style of the trend, “MC X x DJ Y”, is just naming who played which part [reported].

The template has two parameters — who plays, who hypes — and that is a variant machine. Friends, couples, footballers, fighters, racing drivers, film characters and pets all fit the same white studio [reported]. Pets are the sharpest version: a cat at the keyboard looks like a serious producer and a dog makes a great hype man, and for pets AI generation is the only realistic option — a real dog will not point at a laptop on the beat [reported]. A trio of goats made with a CapCut template is reported to have passed 1.2 million plays, and the parrot pair passed 900,000 on one guide’s count [reported].

The second accelerator is the fake-making-of confusion. The sketch has been shared for years as the real story of how the song was made, and the AI remakes inherit that argument: is it real, is it AI, did Jamel really write the song [reported]? The answer — scripted sketch, Stromae wrote it alone — is part of the distribution, because every correction is another comment. Mainstream coverage helped too: the Altman–Amodei version was written up by Fast Company on 7 October 2026 [reported].

The honest clock: the AI remakes started in late September, the template wave followed within a week, and this page ships on 8 November — about six weeks in [reported]. That is old enough for the search term to be real (see the limits section) and old enough for the head term to be crowded: at least eight dedicated template sites and guides hold a version. The durable part is not this sketch, it is the mechanism: one fixed two-person scene, two swappable roles, one song everyone recognises in two notes.

There are three routes, and they are not the same product. The app-template route pastes your photos onto the trend; the dedicated-generator route rebuilds both people in the scene; the generic-generator route films a new scene around your photos and a reference clip. The durations differ by route — 10 to 30 seconds — because they are different cuts of the same sketch, not one canonical clip.

Route 1 — the CapCut template (claimed, no link verified). Guides say Higgsfield and CapCut both have templates for the scene [reported], and one template site says several CapCut and Dreamina template posts for the trend appeared in October 2026 and change often, so to check inside those apps [reported]. No CapCut template link or price for the white-studio rebuild has been verified [unknown]. What can be verified is narrower: one template site warns that the older CapCut items under the same song name are a 7-clip edit and the 2021 group dance — CapCut pastes, not the white-studio sketch rebuilt from two photos [reported] — and one Instagram tutorial from 17 October shows a CapCut code for an “Alors en danse – Halloween edition” variant, a seasonal reskin, not the plain sketch [reported]. If a CapCut result promises the two-photo rebuild, check inside the app whether it regenerates both people or just slides photos over the music before you treat it as the same product.

Route 2 — the dedicated generators (from $4.99 first video). All of these run in the browser: two photos in, MP4 out, no prompt to write. The cheapest verified entry is a first 480p video at $4.99 on the site that sells 15-second and full 16-second cuts at 100–400 credits, from a $9.99 starter pack [reported]. The single-trend site with 10/15/20-second cuts charges $9.99 for one video (50 credits), $22.99 for three and $49.90 for eight, at 480p, with failed renders free and credits that never expire [reported]. The 12-second site charges 10 credits for a standard 720p video, with packs from $6.99 for 10 credits, and states plainly that it is not free and has no free trial [reported]. The site with the human and pet reels charges 168 credits for the 11-second human version and 228 for the 15-second pet version, on Seedance 2.5 at 720p [reported]; its smallest pack is $9.99 for 100 credits, so the dollar price of one video is not stated on its template page [unknown]. The ~30-second site shows 36 credits on its generator and lists 32 credits for 480p/8s and 40 for 720p/8s (about $3.33 at 12 credits = $1), with packs from $9.90 for 90 credits [reported]. A separate site sells the same idea for $8.88 for one ~15-second 720p video, $16.99 for three and $39.90 for ten — but with original AI electronic audio, not the commercial song [reported]. Most of these sites refund failed generations automatically [reported]; check the exact credit price on the generate button before you spend anything, because credits are priced per site, not per video, and the same 168 credits buys a different amount of video on each one.

Route 3 — the generic generators (Seedance, Kling, Higgsfield, Kapwing). Seedance 2.5 is the engine behind most of the dedicated templates [reported], and you can run it yourself with a reference clip plus two photos. The prompt skeleton below is our wording of the circulating structure, not a copy of any single prompt site’s text. Two warnings travel with this route. First, Kling’s own Motion Control guide says that when a motion reference has two or more people, the tool uses only the motion of the person taking up the most space in the frame [reported] — so a one-person motion tool means running it once per person on a cropped clip and stitching the results, and the shared moments rarely line up [reported]. Second, assign the roles by what each person does, never by left or right: the two cross over, and a model told “left plays keyboard” will swap them mid-clip [reported]. Higgsfield’s Genjutsu route is the same idea as a product: a built-in 30-second studio reference clip, two photos assigned by role, 5–30 seconds on Seedance 2.5, vertical 9:16 [reported] — at 720p a 10-second video costs 175 credits on Seedance 2.5 or 32 on Mini, with the cheapest 5-second Mini cut at 8 credits [reported], and its dollar price is not on the page [unknown]. Kapwing’s custom Kai workflow is the no-prompt version of the same route: upload two reference photos, assign keyboard player and animated partner, and the workflow already includes the original audio [reported]; its price was not stated on the guide page [unknown].

Recast the reference clip with the two uploaded photos, shot for shot: the same bright white studio, the same table with the laptop and the mini keyboard, the same light, the same camera angles, cuts, timing and moves.

Image 1 = the person playing the compact keyboard at the table. Image 2 = the expressive partner beside them who claps, points, laughs and leans toward the lens.

Assign the roles by what each person does in the reference, never by left or right screen position. Replace each original performer completely. Keep each person's face, hair, skin tone, build and outfit from their own photo in every shot, close-ups included.

Exactly two people. No original faces, no role swapping, no face blending, no extra fingers, no changing props. No text, no logos.

Sound: most dedicated templates ship the song baked into the MP4 [reported] — which is exactly the copyright problem in the next section. The safer posting path, used by the ~30-second site and the AI-audio site alike, is to export silent or originally scored and add “Alors on Danse” from TikTok’s or Instagram’s own sound library when you upload [reported]. Either way, turn on the platform’s AI-generated label before posting: TikTok requires it for realistic AI video of people [reported].

Photos: one clear, front-facing, evenly lit photo per person, face and upper body visible, no sunglasses, no heavy filters [reported]. The partner’s photo matters most — that face fills the frame in the closing close-up [reported]. The two photos do not need to be taken together or in the same place [reported]. Pets work on most dedicated sites: one animal per photo, whole animal in view, face and markings clear [reported] — but not on every site, so check before you pay: one dedicated site turns pets, cartoons and group photos away [reported].

The search term is real — and already crowded. On 8 November 2026 Google autocomplete returned seven AI-task suggestions for “alors on danse ai video”: the term itself, generator, generator free, free, the misspelling “alors on dance ai video”, “stromae alors on danse ai video”, and “how to make alors on danse ai video” [reported]. The bare “alors on danse” returns song-only suggestions — lyrics, meaning, translation — so the AI demand attaches to the suffixed phrasing, not the song [reported]. But at least eight dedicated tools and guides already hold a version, and the trend is six weeks old, so this page is a guide, not a tool pitch — and it does not expect to outrank the tools on their own name. The room is in the long tail: cost, free, how to make, CapCut, pet version.

One suggested long tail did not check out. “Alors on danse ai video two photos” returned no autocomplete suggestions at all on 8 November 2026 [reported], even though two photos are the defining input of the trend. Searchers ask for the generator and the free version, not for the photo count. Do not build a page on “two photos” as a search phrase; build on it as the explanation, which is what this page does.

The durations do not agree, and that is a real conflict. The same sketch is sold as 10, 11, 12, 15, 16, 20 and 30 seconds depending on the site [reported]. One site even disagrees with itself: its template page lists the human reel at 16:9 and the pet reel at 4:3, while its beat table calls the human frame “almost square” [reported]. Another site’s own blog quotes different credit prices for the same slots than its template page does — 5 credits vs 32 for 480p/8s, 10 vs 40 for 720p/8s [reported]. If a route promises the “original” at a different length or price, check which cut and which price table it is before you pay.

The play counts do not agree either. The parrot reel is 805,700 views by 10 October on one site’s dated count and “passed 900,000” on another guide with no date [reported]. The sketch’s April 2026 repost is “passing 21 million plays” on one site [reported] and two March/April reposts “passed 20 million plays each” on another [reported] — different reposts, different counts, neither independently verifiable from here. One trends listing gives the Alors on Danse trend “≈14M views” with no explanation of what the figure sums [reported]; this page does not repeat it as the views of any single clip. Treat every big number on this page as a dated, sourced claim, not a live count.

The music is a copyright risk if you export it baked in. The footage is Kissman Productions’, the song’s master is Mosaert’s (2009 © & ℗ Mosaert), and its publishing is administered by Warner Chappell [reported]. One guide puts it bluntly: a 30-second clip with the track underneath is not licensed by anyone’s template [reported]. The template sites say the same in their own words: one does not license the song and tells you to add the trend sound in TikTok, Reels or Shorts [reported]; another says a generation purchase covers the tool’s work and grants no license to the song or to a depicted character [reported]; a third says the song belongs to its owners and it gives you no rights to it [reported]. Adding the sound from the platform’s own library when you post is the licensed path; exporting a generated clip with the master baked in and posting it elsewhere is a copyright risk. The one clean workaround on the market is a site that ships original AI electronic audio instead of the commercial recording [reported].

Consent still applies — and the sketch is a fake. Only use your own photo, or a friend’s with their yes; never a child’s or a stranger’s photo [reported]. Most dedicated sites are adults-only for the same reason, and several refuse celebrities, public figures and minors outright [reported]. One site goes further and refuses pets, cartoons and group photos too [reported]. The sketch itself is a scripted parody in which Jamel pretends to have written a song Stromae wrote alone [reported] — so never present a remake as the real making-of, and never present an AI recast of a real person as something they did. Turn on the platform’s AI-generated label before posting [reported].

Two prices and one template could not be verified. The dollar price of a Flowlio video is not on its blog page — only the credit prices, 500 at 480p, 800 at 720p and 1,250 at 1080p [reported] — so no dollar figure is given for it here [unknown]. The dollar price of a Genjutsu video is not on its template page either, only credits [unknown]. And no CapCut template link for the white-studio rebuild has been verified at all [unknown]. All three are marked, not smoothed over.

The failure modes are two-person failures. The two faces blend into one, the roles swap mid-clip, the original performers’ faces show through, and hands on the small keyboard grow extra fingers [reported]. After the 9-second cut the person at the back can drift back to the stand-in’s clothes [reported]. Fixes that recur across the guides: photos that look different from each other, roles assigned by action not by side, “exactly two people” and “no face blending” in the prompt, and re-running — every run is a new take [reported].