Make Any Character Dance: An AI Dance Video Guide
Turn a photo into a dance video that hits every beat. How to choose the dance clip, prep your dancer, cast a duet, go 9:16 and keep the original song.
How AI Dance Videos Work
An AI dance video starts with two things: a clip of the dance and a picture of the dancer you want. The model studies the routine in the clip and generates a new video of your character performing it, step for step.
Genfire does this in Genjudo, its video-to-video studio, which runs ByteDance's Seedance 2.5 reference-to-video model. The preset built for dance is called Dance, beat for beat. It carries over the footwork, the arm timing and the weight shifts of the original, holds the camera framing, and keeps your character's face, clothes and proportions as they appear in your image. Because the result follows the clip's timing, you can keep the original song on it and the moves still land on the beat.
This guide covers choosing a dance that transfers well, preparing your dancer, two-person routines, vertical output for Reels and TikTok, and the soundtrack. For a quick overview of the feature, see the AI dance video generator page.
Step 1: Pick a Dance Clip That Transfers Cleanly
Dance is the hardest kind of motion to transfer, because rhythm lives in small details: where the weight sits, when a heel drops, how far an arm swings past the body. The clip needs to show all of it.
- Head to feet in every frame. Feet carry the beat. If the camera crops at the knees, the model has to guess the footwork, and guessed footwork slides.
- One dancer, or two who stay apart. A group routine where bodies overlap is hard to read. A solo, or a duet with space between the dancers, is easy.
- A steady camera. Most dance trends are filmed on a phone propped against something, which is ideal. Fast handheld moves make it harder to separate the dancer's movement from the camera's.
- Contrast between dancer and background. Light clothes against a dark wall, or the reverse, and nobody walking through the back of the shot.
- One clean phrase. Genjudo's output runs 4 to 30 seconds. One chorus or a couple of eight-counts is usually all a post needs.
You can upload the clip, or paste an Instagram reel link straight into Genjudo and use the reel as the source.
Use dances you have the right to use. That means clips you filmed yourself, clips you licensed, or footage from creators who have said yes. Cast only people who have agreed to appear; your own original characters are the safe default.
Step 2: Prepare Your Dancer
The character image works best when it looks like a casting photo for the routine.
- Full body, feet included. If the clip shows the whole body, the image should too. Shoes visible in the image are shoes the model can keep.
- A plain background. A clean backdrop makes it clear which pixels belong to the dancer.
- Arms slightly away from the body. A neutral, open pose shows the shape of the arms and the cut of the outfit.
- The outfit you want on screen. Pick the look you want to see moving, because that is the look the model will try to hold through the routine.
- The same orientation as the clip. For a vertical dance, a tall full-length image fits the frame best.
If you don't have a dancer, build an original character in the AI Influencer Studio, save a full-body image, and bring it into Genjudo. A character you created can dance in as many videos as you want without borrowing anyone's likeness.
Step 3: Set Up the Run
- 1Open Genjudo and choose Motion transfer.
- 2Drop in the dance clip, or paste the reel link.
- 3Add your dancer image.
- 4Pick the Dance, beat for beat preset.
- 5Leave the prompt box empty to have Genjudo's AI director watch the clip and write a shot-by-shot direction, or write your own.
If you write your own, the clip is @Video1 and your dancer is @Image1:
Transfer the dance in @Video1 onto @Image1 beat for beat: the same footwork, arm timing and weight shifts, with the camera framing held. Keep the identity, clothing and proportions of @Image1 unchanged. Set it in a bright white studio with a polished floor.
The last sentence is optional. Use it when the setting matters, or leave it out and let the direction follow the clip.
Two things are worth spelling out in your own direction. First, anything you want kept: face, outfit, hair, shoes. Anything you don't mention is open to change. Second, any move that matters most, such as "the spin on the final beat" or "the arm wave in the second half". Naming a signature move tells the model where to spend its attention.
Two-Character Dances
For a duet, use the Two characters preset with two images. @Image1 takes the foreground role and @Image2 the one behind, so put your lead dancer first.
Duets need a little more care with the source clip:
- Dancers who stay apart. Partners who cross in front of each other or swap sides make it harder for the model to keep track of who is who.
- Positions named in the direction. "@Image1 is the dancer on the left, @Image2 the dancer on the right" removes the guesswork when the routine has both dancers side by side.
- Two images in matching framing. Both full body, both on plain backgrounds, both showing the outfits you want.
Transfer the dance in @Video1 onto two characters: @Image1 as the dancer on the left, in front, and @Image2 as the dancer on the right, behind. Keep both identities, outfits and proportions, and match the timing, blocking and camera framing.
Going Vertical for Reels and TikTok
Reels, TikTok and Shorts are 9:16. Set the output to 9:16 before you run, and start from a vertically filmed clip where you can. A dance shot on a phone in portrait already frames the full body for a tall screen, which is the simplest case for the model: nothing has to be invented above the head or below the feet.
The same logic applies to your dancer image. A tall, full-length image matches a tall frame. A square crop that cuts the character off at the shins asks the model to invent the feet, which is exactly where dance needs the most accuracy.
Keeping the Original Song
Under sound, choose Clip's sound and Genjudo keeps your source clip's audio on the finished video. Since the dance follows the clip's timing, the steps stay with the song.
Three tips:
- Match the length. Set the output length to the length of your source clip, so the music doesn't stop early or the dance doesn't run past the last bar.
- Want different music? Pick AI sound for a soundtrack Seedance writes for the result, or Silent to add your own track, captions or voiceover later in the AI video editor.
- Check the rights to the song. The music has an owner too. Keep the original track only when you're allowed to use it.
Fixing a Routine That Doesn't Land
| What you see | Likely cause | Fix |
|---|---|---|
| Feet slide instead of stepping | Feet cropped in the source clip | Use a clip with the full body in frame |
| Moves look softer than the original | Fast handheld camera | Pick a clip shot on a steady phone or tripod |
| Dancers trade identities mid-routine | Partners cross or overlap | Choose a duet where the dancers stay apart, and name positions in the direction |
| The outfit changes during the dance | Outfit hidden or cropped in the image | Use a full-body image that shows the whole outfit |
| The video stops before the routine does | Output shorter than the clip | Raise the length, or use Extend to continue the result past its last frame |
Run two to four takes for a routine you care about; Genjudo allows 1 to 4 per run. Generative video varies between takes, and a fast routine has more room for variation than a walk.
When to Try Motion Control
The Motion Control tab in Genjudo runs Kling 3.0 Motion Control, a different model that moves a character image the way a reference video moves. If a routine isn't landing in Motion transfer, run the same clip and image through Motion Control and compare the two. The Kling V3 page covers the rest of that model family. If you are choosing between dance apps, our Genfire vs Viggle comparison is a good place to start.
What It Costs
Genjudo bills per second of output, and every take is billed for its own seconds. It all comes from one Genfire credit balance shared with 80+ image, video and audio models, and the exact cost is shown before you run. Credit packs start at $19 on the pricing page.
FAQ
How do I make a photo dance with AI?
Open Genjudo, choose Motion transfer, add a dance clip and a full-body photo of your character, and pick the Dance, beat for beat preset. Seedance 2.5 generates a new video of your character performing the routine with the same footwork, arm timing and camera framing.
Can I keep the original song on my AI dance video?
Yes. Choose Clip's sound and Genjudo keeps your source clip's audio on the result. Only do this with music you have the right to use.
Can two characters dance together?
Yes. The Two characters preset takes two images and casts one in the foreground role and one behind. Source clips where the dancers stay apart give the cleanest results.
What size should a TikTok or Reels dance video be?
Use 9:16 vertical. Set the output to 9:16 and start from a vertically filmed dance clip, so the full body already fits the tall frame.
How long can an AI dance video be?
Each result runs 4 to 30 seconds. To keep a routine going past the end of a result, use Extend to continue it beyond its last frame.