MJ Dance is a one-tap file, not a prompt you rewrite weekly

mj_dance is a photo in, a clip out. A new generate is a different object. The template id is mj_dance. The page is MJ Dance. Upload one still — the required field is user_image_url, labelled "Your photo" — and the format already chose the motion: restyle the look and dance like Michael Jackson, driven by the MJ reference clip. You do not rewrite a Veo paragraph every Monday to keep the moonwalk.

Estimated cost is 71 credits, about 78 for a full-quality render. That figure is listed before you run. A blank generate with a new dance prompt is not cheaper because you typed it; it is a different SKU, a different model choice, and a different chance the subject is no longer the person in the photo.

The creative choice already happened

Run a one-tap AI video template is the named agent job. start_template_run kicks off after every required input is a hosted URL, usually a photo attached with the chat + button. get_template tells the agent which inputs this template needs. A live card tracks get_template_run; cancel_template_run exists if you need to stop. You do not poll in most cases. You do not write a scene list.

No timeline. No "make it more Thriller this week." The motion structure is locked in the template. Your job is the still: a sharp, well-lit photo the identity can survive. A screenshot of a screenshot is how the dancer becomes a cousin.

The templates catalog is the shelf. Finger snap, mugshot, Mini Me, MJ Dance — each is a format, not a prompt pack. Rewriting weekly is how you accidentally leave the shelf and start paying for movies you did not storyboard.

A generate is not a cheaper template

If you wanted to pick Veo versus Kling per shot, you wanted a generate or a movie, not one-tap. The template already chose its stack. Asking the agent for "MJ dance but make the background Tokyo and change the choreography" is a new brief stuffed into a template name. The template does not take a scene list. It takes the photo.

If no template matches, generate. Forcing MJ Dance to "just be a product ad" is how you pay ~71 credits for a joke format you will not post.

One-tap is also not motion transfer. Motion control needs a character still and a driving clip you supply. MJ Dance already owns the driver. You do not attach a second dance video. You attach a face. Missing that distinction is how people paste a prompt, skip the photo, and wait for a moonwalk that never had a body.

Cloning a public workflow is a third object: saved, not auto-run, a scene graph you can rewrite. start_template_run is the finger-snap class: photo in, clip out.

Run it, or don't. Don't maintain it.

The test: is there a named one-tap template, and do you have the photo it requires? If yes, attach, run, watch the card. If you needed a new plot around locked characters, that is a saved workflow. If you needed a custom dance from a clip you own, that is motion transfer. MJ Dance is none of those. It is one file in, one file out, listed credits, no weekly prompt hygiene.

FAQ

Do I pick the video model inside MJ Dance?

No. The template already chose its stack. You supply the photo. If you wanted to pick a model per shot, you wanted a generate, not one-tap.

Can I skip the photo and describe the dancer?

No. user_image_url is required. A paragraph about a performer is a different generate, and a worse one when identity is the point.

Is rewriting the prompt how I get a new look next week?

That is a new generate, a different object. A new still in the same template is still MJ Dance. A new paragraph in a text-to-video box is not. Do not maintain a prompt file for a format that does not take one.

What if the run is in progress and I hate the still?

cancel_template_run exists. Closing the chat is not a cancel. Starting before the photo is a hosted URL is a failed run you still have to notice. Attach first.