How to make an AI dog dance video from your own photo

By Vellm · October 19, 2026

You can make an AI dog dance video from one clear photo. In Vellm, choose a dance template if it already has the routine you want, write your own routine in Create, or remake a trending dance clip with your dog as the performer.

The useful first decision is the kind of clip you want. A dog doing a recognizable routine needs a different starting point from a dog gently swaying in your living room.

Pick a routine before choosing the photo

Watch the template preview from start to finish. Check whether the dog dances alone or alongside other performers, and see how much of its body remains in frame. A close-up that looks great in your camera roll might not give the model enough visual information to animate a full-body dance.

Vellm currently offers these dog-friendly options:

Each of these templates is a configured remake: a reference dance clip and a tested instruction are already set up, so you only add the photo. These templates export silent video. A song playing in a reference preview does not mean that song comes with your download. Preview labels also matter: a trend reference shows the intended scene, while an actual generated example shows a particular output.

Use a photo that shows the whole dog

Choose a sharp photo with the face, ears, body and paws visible. Aim for even light and a simple background. Keep one dog in each photo so the subject is clear.

You do not need to make your dog stand upright or perform a trick for the camera. Photograph it in a comfortable, ordinary pose. The model creates the dance.

If your best photo crops off the feet, look for another one before generating. Writing "show the paws" cannot recover an exact view that the photo never supplied. For more help, use the photo selection guide.

Make the template version

1. Open the template and watch its preview through to the end.
2. Upload your dog photo into the named slot. For a group template, read the role labels before adding more photos.
3. Check the listed duration, format and credit cost, then follow the permission and generation steps on the page.
4. Watch the completed clip all the way through before downloading it.

For La Calabaza, one photo supplies all three performers. With two photos, the first pet takes the center and the second appears on both sides. A third photo lets you choose the right-side pet separately. Adding photos changes who appears, so do not fill the slots with arbitrary pictures of the same dog unless that is the result you want.

Describe your own dance in Create

For an original scene, open Create, choose Animate a photo, and upload your starting image. Select a model supported in that mode and begin with a short, simple action.

Try this motion prompt:

The dog sways gently from side to side, lifts one front paw, then sets it down. Its tail wags. A locked camera keeps the whole dog in view in one continuous shot.

This is a starting prompt, not a tested promise of a particular result. A complex human dance may change the dog's anatomy or markings. Review the first attempt before asking for faster movement, spins or extra performers.

Write a second-by-second routine

Once a simple prompt works, you can try planning the whole clip. Split it into short time windows and give each window one main move. This guides the order and rough timing, instead of leaving the model to pick moves on its own.

Here is a 10-second routine for Animate a photo:

One continuous shot. The camera is locked at the dog's eye level and the whole dog stays in frame, paws included. Keep the room from the photo as the background.

0–2 s: The dog sits up on its hind legs and lifts both front paws to chest height.
2–4 s: Still upright, it sways its body to screen-left, then to screen-right, ears bouncing.
4–6 s: It drops back onto all four paws, turns once in place and faces the camera again.
6–8 s: It hops twice on the spot, lifting its screen-left front paw high on each hop.
8–10 s: It sits, tilts its head toward screen-right and wags its tail. It holds this pose to the end.

Keep the dog's breed, face, ears, coat colors, markings and body size from the photo throughout. Its front legs end in real dog paws with pads, never human hands or fingers. No extra legs, no other animals or people. No text, captions, logos or watermarks.

A few rules make timed prompts easier for the model to follow:

• Set the duration control to the last timestamp. A 10-second timeline in a 5-second clip gets compressed or cut off.
• Start with windows of about two seconds and one move each. Every extra move in a window is something the model may drop or blend.
• Use "screen-left" and "screen-right" for directions. "Its left paw" is ambiguous once the dog turns.
• Describe the body, not the song. "Sways on the beat" means little in a silent clip; "sways to screen-left, then screen-right" does.
• Keep the appearance and anatomy lines. They help the model preserve the coat and paws during fast moves, which is where AI dog videos most often go wrong.

Treat the timestamps as a plan, not a guarantee. When we built Vellm's dance templates, a written timeline alone did not reliably keep exact timing or full-size moves. A reference clip did much better. If you need a specific existing dance, remake it instead.

Remake a trending dog video

If the dance you want is already a popular clip, show it instead of describing it. In Remake a video, the clip supplies the moves, timing and camera, and your photo supplies the dog. Use clips you have permission to use.

1. Trim the clip to the dance itself. Cut intros, reaction shots and end cards. A clip with one clearly visible dancer is the easiest start.
2. Open Create, choose Remake a video, and add the clip and your dog photo. The clip becomes @Video1 and the photo @Image1.
3. Check the settings. The video's length follows your trimmed clip, and the aspect ratio is set to Auto by default.
4. Write a description that tells the model what to take from each file. A remake needs one, so once both files are in, Vellm fills in a one-line example. Replace or extend it with the prompt below.

This is the structure Vellm's own dog dance templates use:

@Image1 is my dog and the only animal in the video. Keep its breed, face, ears, coat colors, markings, fur length and body size from @Image1 throughout. Do not copy the look of the dog in @Video1. Use @Image1 only for the dog's appearance, not its background.

Follow @Video1 for the whole routine: the order of the moves, their timing and size, the dog's position in the frame and the camera movement. Do not slow down, shrink or skip any move. Keep the setting and lighting of @Video1.

The front legs end in real dog paws with pads, never human fingers. No extra legs, people or other animals. No text, captions, logos or watermarks.

Expect to iterate. A short prompt like this is a starting point, and the first result may soften a move, skip a step or drift from your dog's look. Vellm's own template prompts went through several revisions, and they also spell out the routine move by move. If moves go missing, watch the clip and add its key moves in order, with rough times:

Key moves in @Video1, in order: 0–3 s the dog stands on its hind legs and waves both front paws. 3–6 s it turns side-on to the camera and lifts one hind leg. 6–9 s it walks toward the camera until its face fills the frame. Keep every move at full size.

Change one thing per attempt, such as adding the move list or trimming the clip, so you can tell what helped. Each completed generation uses credits, even if you decide to adjust it.

Adjust the main prompt to the clip you are remaking:

• The reference dog looks very different from yours. Name the difference: "Keep my dog's long legs, not the short legs of the dachshund in @Video1." Without this, the model may copy the reference dog's build.
• The trend features a person, not a dog. Add this to the second paragraph: "My dog takes the place of the dancer in @Video1, dancing on its hind legs. Hand gestures become paw gestures."
• Several performers are in the clip. Say which one your dog replaces, for example "the dancer in the center", and what happens to the others: keep them as new people, or remove them.
• You want a different setting. Replace "Keep the setting and lighting of @Video1" with your own scene, such as "a sunny backyard lawn".

Type @ in the description and pick the file from the suggestions, so the labels match your file cards. See how to point to uploaded files and the full remake guide for trimming and review tips.

A remake follows the movement, not the soundtrack. If you want the trend's music, add a track you have the rights to use when you edit or post.

Check the clip and the cost

Look at the ears, face markings and paws as the dog turns. Watch for extra limbs, sliding feet or a body that changes shape. If the face is already unclear at the start, replace the photo. If it changes only during a fast move, simplify that movement in Create or choose a less demanding routine.

A 50-credit template can be covered by the $4.49 short pack. The standard 100-credit pack costs $7.99 and covers 2 50-credit attempts or one 100-credit template. Those are USD pack prices, not a guarantee that you will keep every generated attempt. Check current pricing before buying.

Template downloads are silent. Create has an Audio control on supported models, so check its setting separately. Add music available for your intended use when you edit or post the clip.