Free prompt

Smalltown Boy AI prompt

The Smalltown Boy AI dance is a reference-video job, not a text-to-video one: a prompt alone will not reproduce the moves or the cut. You need the original clip as the motion reference, one photo per performer, and a model that does motion transfer (Kling, Seedance, Higgsfield and similar). Below is the setup that works for us.

What to prepare

1. The original duo clip, 14 seconds, 9:16 vertical, as the reference video.
2. Photo 1: whoever takes the close-up first. Photo 2: whoever dances first. One person or pet per photo, facing the camera.
3. A first frame with both subjects already in the scene. Models keep faces much better when the video starts from it.

Step 1: the image prompt (first frame)

For an image model that takes several reference images (GPT Image, Nano Banana, Seedream). Image 1 is photo 1, image 2 is the first frame of the original clip, image 3 is photo 2.

Use the first frame of the Smalltown Boy duo dance as the scene (image 2). Put the subject from image 1 in the close-up: leaning toward the camera, both hands on the head, mouth open as if screaming along. Put the subject from image 3 a few steps behind, mid-dance. Keep each subject's face, hair or fur and clothes from their own photo. Keep the red fence, the light and the vertical framing of the scene.

Step 2: the video prompt (motion transfer)

Reference video: the original clip. Reference image 1: the first frame from step 1. Reference images 2 and 3: the two photos, in the same order.

Follow the motion of the reference video exactly. The subject in reference image 2 performs the close-up part for the first 7 seconds and then the dance part; the subject in reference image 3 dances first and takes the close-up after the cut. Keep both subjects' faces and clothes the same in every shot. Start from reference image 1. No extra people, no text, no camera moves that are not in the reference.

Where it usually breaks

• The dancer in the background of the first shot stays the original person. Name both roles in the prompt and give the first frame.
• The two subjects swap faces after the cut. Bind each photo to a role, not to a position on screen.
• The sound. Most video models return silent or new audio; put the original track back and trim the video to the clip length.
• Pets. They need to stand upright; photos where the pet faces the camera work best.

Or skip the prompt

This site runs exactly this pipeline for you: two photos in, a free preview of the first frame, then the video with the original sound. $6.90 for 480p, $10.90 for 720p. No account, no app, no prompt.