How can my character follow a dance video?
To make your character perform an existing dance, provide a clear character image and a video of the routine, then use a workflow that accepts both.
This AI dance video workflow uses the image for appearance: face, body, clothing or fur. The clip supplies a performance to follow. When choosing an AI dance generator, check that it accepts a video reference. Writing “make this character dance” with only a photo leaves the choreography to the model. If you want a particular arm wave, turn or finish, use the routine’s video rather than relying on its name alone.
Choose by the result you want. Photo animation from a text prompt can invent a dance. A workflow accepting a motion video can follow an existing performance. Face-only replacement aims to keep the source body and clothes; whole-person replacement uses your character’s body and outfit too. Alibaba Cloud distinguishes motion-driven animation, character replacement and broader reference generation. Our BAD example uses reference generation with Wan 3, without a pose rig or a promise of frame-by-frame copying.
In FlyAIgh, the route depends on your dance source. For the prepared BAD routine, open the template and put your image in the dancer slot; the template supplies the dance video. For a clip you recorded or selected yourself, open the own-video tool and provide both the clip and character image. Use whole-person replacement when the new body and outfit matter.
AI cat dance example: BAD from mannequin to human to cat
This comparison follows one 15-second BAD dance segment through two generations. On the left is the green-screen mannequin source. In the middle is our human remake in an office. On the right is the tabby result generated through the BAD template using that office remake as its video reference.
The order matters: the cat run did not use the green-screen mannequin directly. The template’s actual input is the office sample, while its reveal preview includes mannequin footage to show the transformation. A preview is not always the video sent to the model.
You can see a similar dance progression and the office setting in the cat result. The performer’s proportions and clothing change, and matching the broad sequence does not prove identical joints or beat timing. These are two selected outputs, not repeated trials or a model ranking. Watch the entire comparison, especially the arm changes and ending, before deciding whether that degree of motion similarity suits your project.
What character image and dance clip do I need?
A headshot describes the face, but it does not specify the body, outfit or legs. For a full-body character dance, choose an image showing those details, with the face still readable. If you only have an avatar, prepare a full-body version matching the design you want before generating the dance. A character sheet can show additional views; it is still one uploaded image, not a rig with separate limb controls.
- Keep one intended character in the image. If it is embedded in a poster or busy scene, use the workspace’s reference check and extraction option when offered; inspect the extracted image before accepting it.
- Choose a dance reference with the performer and relevant limbs visible. A hand hidden behind the body or feet outside the frame leave less information to follow and evaluate.
- Prefer a continuous passage for the first attempt. Cuts, fast camera changes and multiple dancers make it harder to tell whether the same target and routine were followed.
- Keep the reference framing you need. BAD is a portrait 9:16 template. A horizontal article player can show several portrait clips side by side without cropping away the feet.
These are preparation and review criteria, not measured success-rate claims. If the character has short legs, a tail or a very different body shape, allow for reinterpretation and pay particular attention to balance and contact with the floor.
How do I make my character dance BAD?
Choose this route when you want the routine shown in our sample and do not need to supply your own dance clip.
- Open the BAD template and play the office sample. Confirm that its routine and portrait framing are what you want.
- Add your character image to the dancer slot. Check the selected reference, especially if the workspace extracted it from a larger image.
- For the first run, leave the optional background reference empty. This asks for the office scene to remain and gives you fewer requested changes to inspect.
- Choose an available model and resolution. Wan 3 produced the cat example here; this is a case record, not evidence that it is best for every character. Models that cannot process the template’s duration are unavailable.
- Read the prompt preview and credit quote before submitting. The template already describes the dance and target; a long description of every move is not required.
- When the result is ready, compare the whole clip with the sample. Download the version you want to edit further, then add your soundtrack.
Changing the background at the same time is possible, but it adds another result to judge. First establish a character-and-motion version you are happy with; then make a scene variation. Leaving the background slot empty is a preservation request, not a pixel-level guarantee.
Where do I upload my own dance video?
If the movement you want is in your own recording, use Recast’s upper person and face swap tool. Upload the footage, select the dance passage, identify the performer and choose whole-person replacement to use the character’s body and outfit. Face-only replacement serves a different goal: keeping the original body and clothes.
The selected window can be up to 30 seconds, but the chosen model determines how much can be processed in one run. Longer windows may need segments. Check that the character, camera and final pose remain continuous at each join; accepting separate clips does not automatically make a continuous dance.
Read the complete person-replacement guide
Does the result follow the moves and beats?
Exact choreography and beat timing are not guaranteed. A video reference gives the model a performance to follow; a matching face or similar sequence does not establish identical poses. Watch source and result at normal speed, then compare the same timestamps around important arm changes, turns, foot plants and the ending. Decide which differences your project can accept before generating another variation.
| Check | What to compare | What to try next |
|---|---|---|
| Sequence and rhythm | Arm wave, direction changes and final pose occur in the intended order and near the intended beats | Recheck the selected passage; try a shorter continuous section if timing drifts |
| Body and identity | Face, clothes or fur stay consistent through turns and occlusion | Inspect the reference image’s missing views and the target description |
| Hands and feet | Limbs remain readable, feet stay in frame and floor contact looks plausible | Check visibility in the source and full-body coverage in the character image |
| Scene and joins | Background changes are acceptable; segmented clips connect without an abrupt pose or identity jump | Reduce simultaneous changes; review each segment before assembling the dance |
If the face looks right but the dance does not, first locate the mismatch at the same timestamp in both clips. Check the selected passage and intended performer, whether the source shows the relevant limbs, and whether the character image specifies the full body. Change one input or instruction at a time so the next result is easier to compare. These checks do not diagnose the cause of a particular failure or guarantee a fix. A prompt requesting exact motion is an instruction, not a measurement.
Check the visuals before choosing music. BAD templates do not include the original song. Add your own recording or a track you are allowed to use in your editor, and compare the key moves with its beats. For uploaded footage, the own-video tool retains source audio by default; that audio option is separate from the template workflow. Browsing is free; generation requires a paid account and credits. Use the quote shown for your selected settings.
Example records and sources
The BAD office remake and template cat output were recorded on 2026-10-09. The cat run used the 15-second office sample, Wan 3, a character reference, portrait 9:16 framing and 720p output. The comparison reuses those files, trims them to the shared 15-second passage and removes audio; no new generation was commissioned for this article. The original performer’s private photo is not published.
Alibaba Cloud: video generation, motion-driven animation and character replacement
That documentation explains the task distinctions; it does not establish FlyAIgh’s available models, limits or prices. The steps above follow the current Recast workspace. This guide addresses following an existing dance; our person-replacement guide covers choosing replacement scope and handling footage more broadly.
Frequently Asked Questions
How do I make my own character dance with AI?
Use a workflow that accepts both a character image and a dance reference video. The image describes the character; the video guides the movement. In FlyAIgh, use the BAD Recast template for a ready-made dance, or the own-video person replacement tool for your footage. Review the complete result before adding music.
Is asking a photo to dance the same as copying a dance video?
No. A prompt such as “dance energetically” asks the model to invent movement. Following an existing routine requires a video reference or a motion-driven animation workflow. Even then, exact poses, timing and foot placement need checking.
How do I make an AI cat dance video?
Open the BAD template, add a clear full-body cat character image to the dancer slot and compare the generated clip with the office sample. Our tabby example follows that workflow; it does not guarantee identical timing for every cat. Check the paws, balance and tail as well as the face.
Can AI preserve every dance move and music beat exactly?
Exact poses and beat timing are not guaranteed by this workflow. Compare source and result at the same timestamps for arm changes, turns, foot contact and the ending. The BAD comparison shows a broad dance progression, not measured frame-by-frame accuracy. Adding music afterwards does not correct movement drift.
Is a headshot enough to make my character dance?
A headshot supplies facial appearance but leaves the dancing body and outfit unspecified. For a full-body character dance, prepare an image showing the intended body, clothes, arms and legs with a readable face. We have not run a controlled headshot-versus-full-body test, so this advice describes input information rather than a measured success rate.
The face looks right, but the hands, feet or rhythm are wrong. What should I check?
Locate the mismatch at the same timestamp in the source and output. Check the selected dance passage and target performer, limb visibility in the video, and full-body information in the character image. Change one input or instruction at a time and compare again. These are review steps, not a diagnosis or a guaranteed repair.
Does the BAD template include the original song?
No. Recast templates do not include the original song. Add your own recording or a track you are allowed to use after checking the generated dance. The own-video tool keeps source audio by default, which is a separate workflow.
Can I make a longer dance, and is generation free?
The BAD template is 15 seconds. For your own footage, select a window of up to 30 seconds; model limits can require multiple segments. Check identity and motion at the joins. Browsing examples is free, while generation requires a paid account and credits. Read the displayed quote before submitting.