Jisoo Addicted AI: How the Viral Dance Trend Works, and How to Join It the Right Way

It is two trends wearing one name. The first is human: "ADDICTED", Jisoo's second pre-release single ahead of her first full solo album, dropped on October 30, 2026 through her agency BLISSOO — a dark dance-pop cut that broke from her earlier solo style, with choreography built by Leejung Lee, the choreographer famous from Street Woman Fighter. The official dance performance video followed on November 5 and trended #1 on YouTube worldwide, appearing in the trending charts of 38 countries. Covers, challenges and duet edits multiplied on TikTok within hours.

The second trend is synthetic: fans noticed that a modern photo-to-video model, given one clear photo and a choreography prompt, produces a surprisingly watchable K-pop dance clip — and the "Addicted AI dance video" became a format of its own. You do not film anything. You supply a photo of a person (yourself, in the responsible version of this trend), describe the routine, and the model animates it to fit the trend's formats.

Ten days took the song from release to worldwide #1 to an AI-video format. The timeline, condensed:

The ADDICTED era look — star-embellished black, long nails, sequin backdrop — that most AI recreations try to copy. Image credit: BLISSOO.

Every version of the trend runs the same pipeline: one photo, one motion prompt, one image-to-video generation. The photo supplies identity and wardrobe; the prompt supplies choreography, camera and mood; the model animates the person in the photo. Three formats dominate:

• ▶

  The mirror solo

  One dancer, medium-full shot, practice-room or hallway backdrop, looping eight-count choreography. The easiest format — one subject, one camera, no interaction.

• ▶

  The hallway duet

  The format the ROSÉ clip made famous: two dancers face each other in a narrow corridor. AI versions feed two photos and ask the model for mirrored choreography — twice the subjects, twice the failure rate, but the most shareable result.

• ▶

  The challenge cam

  Phone-camera framing, slight high angle, the hook move of the routine repeated twice. Built for the official challenge audio, so it slots straight into the trend feed.

• 1

  Prepare the photo — of yourself

  Full-body or three-quarter framing, limbs visible, even lighting, plain background. This is the single biggest quality factor: the model animates what it can see, and a cropped or dim photo produces a stiff dancer. If your photo needs cleanup, our image to image tool can relight or restyle it first — the star-dress look of the era is one prompt away if you want it.

• 2

  Write the choreography prompt

  Describe the routine like a director: which moves, which direction the body faces, what the camera does, what stays fixed. Keep it to one eight-count and say "loops smoothly". Names of real artists do not help the model — descriptions do.

• 3

  Generate with an image-to-video model

  On VidSeedance, run your still through image to video with Seedance 2.0 or 2.5, Veo 3, or Kling. Five to eight seconds is the sweet spot: long enough for a full hook, short enough for the model to hold the body together.

• 4

  Review, fix, repeat

  Judge the clip on three things: does the body stay consistent, do the hands survive, does the loop close cleanly. Re-roll rather than extending a broken take — a failed run does not spend credits here, so iterating is cheap.

The industrial set of the ADDICTED visuals. Naming a set's mood — "dark machine room, green practical lights" — in your prompt gets you closer to the trend's look than naming the artist does.

These are written for the mirror-solo format and a photo of yourself — adapt the wardrobe line to your photo:

The ADDICTED visual world swings from machine rooms to carnival booths. When prompting, pick one mood per clip — blending both in a single take is how AI dance videos turn muddy.

• Simplify the routine. Fast overlapping choreography is where hands and joints break. One clean eight-count beats a failed thirty-second routine.
• Lock the camera in words. "Fixed medium-full shot, no camera motion" removes the biggest source of wobble.
• Describe the wardrobe you actually have. The model blends the photo with your prompt; conflicting outfit descriptions cause flicker.
• Generate short, then extend. A stable six-second loop can be extended or edited together; an unstable long take cannot be repaired.
• Judge takes by hands and loop. Re-roll the failures instead of polishing a broken take — generation is the cheap part.

The fastest way to kill a trending clip is to get it removed — or worse, to have the artist's fandom report it into oblivion. Four lines to respect:

• Be the dancer. Put yourself (or someone who agreed) in the video. Generating Jisoo's own likeness — or any real person's — without permission is impersonation, it violates every major platform's synthetic-media rules, and K-pop fandoms are exceptionally effective at reporting it.
• Ride the official sound. Add ADDICTED inside TikTok, Reels or Shorts from the platform's music library instead of baking the track into your file — that keeps you inside the licenses the platforms already negotiated, and inside the trend feed.
• Label it as AI. Turn on the platform's AI-content toggle. Undisclosed synthetic media is the specific thing these policies punish.
• Monetization needs real permission. Posting to the trend is one thing; running it as branded or monetized content needs the rightsholder's and choreographer's actual sign-off.

The Addicted AI trend is the clearest sign yet that dance content has become a photo-to-video genre: one photo in, a choreographed clip out, no camera and no dance studio. The format rewards simplicity — clean eight-counts, locked cameras, short loops — and the account that survives the trend is the one that puts itself in the video, rides the official sound, and labels the AI honestly.

The song is ten days old and the trend format is still forming. Whoever posts a clean, honest version this week is early; that window will not stay open long.