The SaWaDiKa AI video is a 14-second clip of you doing the jump-cut dance from LISA's 2026 single SaWaDiKa, made from one photo of you. You upload the photo, pick a location, press Generate and download an MP4 with the song already on it.
Key takeaways
- You need one sharp, full-body photo; there is no prompt to write and no routine to learn.
- The clip keeps the original timing and jump cuts, and you wear the one outfit from your photo in every shot.
- The song is mixed in after the render, so the dancer does not sing or lip-sync.
- We have not tested every move, so the choreography can differ from the real dance in places.
- Our trend radar marked the search terms as fading on 2026-11-01, so this is a late entry, not a peak one.
In this guide
- What exactly do I get from the SaWaDiKa AI video?
- Which song and which original dance is it based on?
- How big was SaWaDiKa when it came out?
- Is it too late to post one now that the trend is fading?
- What kind of photo should I upload?
- Why do I dance in the clothes from my photo?
- Which location should I pick?
- Does the video have the song, and will I lip-sync?
- How closely does the dance match the original?
- What does one video cost, and what if it fails?
- Is videosi connected with LISA or her label?
What exactly do I get from the SaWaDiKa AI video?
You get a vertical or widescreen MP4, 14 seconds long, in which a dancer who looks like you performs the SaWaDiKa routine. The original cuts between a sunny rooftop with a city skyline and a wet train platform, and the SaWaDiKa AI version keeps that rhythm of cuts if you choose the original location.
Under the hood the page uses a motion reference clip for the moves and your photo for the person. The instruction swaps the original dancer out entirely, so her hair, tops, trousers, gloves and boots are meant to disappear rather than blend into yours. That is a design choice we made because anything left unnamed tends to stay on screen.
Which song and which original dance is it based on?
The song is SaWaDiKa by LISA. The title is her spelling of the Thai greeting sawatdee ka, and Billboard Canada reports the single arrived on Friday 4 September 2026 with a high-fashion video shot in Bangkok, with multiple outfit changes and scenes on rooftops and in an underground tunnel.
The short dance clip that spread on social platforms is a separate piece of footage from the music video. It runs under fourteen seconds and cuts between two places, and that is the routine we rebuilt. LISA also put out a full dance performance video later, which Soompi dates to 12 September 2026 (KST). That one features formations with backup dancers, and our version does not: it always has one adult dancer.
How big was SaWaDiKa when it came out?
Very big. Bandwagon reports 70.8 million YouTube views in the first 24 hours and 100 million in just over 48 hours, the fastest music video to reach that mark in 2026.
Those figures belong to the music video, not to AI clips. We do not have a view count for any AI version and we are not going to guess one. What we can say is that people started swapping themselves and characters into the routine in October 2026, and that the interest we track has already started to cool.
Is it too late to post one now that the trend is fading?
It depends on what you want the clip for. If you are chasing views from a search wave, the peak has probably passed: our radar first caught the terms on 14 October and marked them as fading on 2026-11-01. If you want a personal clip for friends, a group chat or a dance-themed edit, timing matters much less.
New covers were still appearing on the day we checked, so the format is not dead. The SaWaDiKa AI page stays up as long as people keep looking for it, and the cost of a single try is small enough that waiting for a perfect moment is not necessary.
What kind of photo should I upload?
Use one sharp, well-lit photo that shows your face and your whole body, from head to shoes. This is a full-body dance, so a small or blurry face has few pixels to hold on to across 14 seconds of fast movement.
Avoid sunglasses, hats and heavy filters, and avoid photos cut off at the waist. If the photo stops at your hips, the model has to invent your legs and shoes, and that is where the result stops looking like you. Include only one person; the clip is built for one adult dancer.
Why do I dance in the clothes from my photo?
Because the source clip changes outfit at each cut, a model left alone will tend to change yours at the cuts too. We tell it to dress you in the single outfit from your photo in every shot, so the clothes you upload are the clothes you dance in, even when the location jumps.
The practical consequence is that you cannot ask for the rooftop costume or the red jacket from the original. If you want a different look, change the photo. Also note that we do not try to make you look like LISA; the aim is you, doing her routine.
Which location should I pick?
There are three. The original skyline and platform keeps the two places and the cuts between them, which is the closest to the real clip. Night street puts every shot at an empty city crossing with lit shophouses and string lights. Dance hall puts every shot in a dark hall with a polished floor and a small mirror ball.
The two new options use one place for all cuts, so the jump cuts become changes of angle rather than changes of scene. Pick the original if you want recognition, and one of the others if you want the clip to look like your own take.
Does the video have the song, and will I lip-sync?
The finished video carries the song. The audio from the reference clip is mixed in after the render, so the file you download already has it. Nothing is sung by the AI dancer: the mouth stays closed or relaxed, and there is no lip-sync.
That also means you can post the MP4 as it is. If you prefer a different track, replace the audio in any editor, since the file is an ordinary MP4.
How closely does the dance match the original?
We have not tested every move against the real routine, so treat the result as a close imitation and not a frame-perfect copy. The instruction asks for the same moves and timing: palms together above the head at the start, sharp arm lines and hip hits, a knee lift, hands framing the face, and a final jump with one arm thrown up.
The weak spots are fast arm movements, where extra fingers or soft hands can appear, and the cuts, where details of your face or outfit can drift. If a take looks off, try again with a sharper photo before changing anything else. We would rather say the dance may differ from the original than promise a perfect match.
What does one video cost, and what if it fails?
A 14-second video on the Standard model at 720p costs 12 credits, and an account's first one costs 10. Better models and higher resolutions cost more. Credit packs start at $6.99 for 10 credits and never expire, a Trend Pass is $9.99 for 30 credits, and plans start at $29.90 a month for 100 credits.
Nothing is generated without credits, and there are no refunds on purchases. If a render fails, its credits come back to your account automatically. The SaWaDiKa AI page shows the exact cost for your chosen model before you press Generate.
Is videosi connected with LISA or her label?
No. videosi is an independent tool and is not affiliated with LISA, LLOUD, BLACKPINK or Sony Music. The song, the choreography and the names belong to their owners, and we only make a personal dance clip from a photo you provide.
Use photos of yourself or of an adult who agreed, and label the result as AI-generated when you post it. Do not use the clip to suggest that LISA or her team took part.
