Smalltown Boy AI Video Prompt: Higgsfield, Kling & CapCut

A useful Smalltown Boy AI video prompt describes two distinct subjects, the foreground/background exchange, and the final dance together. The song title alone does not describe those movements. Use the original prompt below as a starting point, then choose the workflow that matches your source material.

If you want a prepared two-photo workflow, open the Smalltown Boy AI Dance Generator.

Copy this two-person dance prompt

Create a vertical two-person comedy dance in a simple, evenly lit setting. Subject A is [describe the first person or pet]. Subject B is [describe the second person or pet]. Keep their faces, clothing or fur markings distinct and consistent throughout. Start with A in the foreground making an exaggerated surprised expression, hands near the head, while B dances farther behind. Let A move back and B move forward for the second reaction. End with both subjects fully visible, dancing side by side with small coordinated steps and relaxed arm movements. Use a fixed wide-angle composition, enough space between the subjects, stable lighting and a continuous background. Preserve clear hands, separate limbs and recognizable identities.

Replace the bracketed descriptions. For cats, specify each coat pattern; for people, distinguish clothing and hairstyle. If your tool provides a negative prompt field, try “merged bodies, identity swaps, extra limbs, cropped feet, sudden camera cuts, flickering background.” These are suggestions, not guaranteed controls.

Choose the right workflow

• Higgsfield Genjutsu — A motion reference and subject references; Carrying motion into a different cast or setting
• Kling image-to-video — A clear starting image and a text prompt; Making an original interpretation of the sequence
• CapCut editing — Recorded or generated clips; Adjusting pacing, captions, music and the final edit

This is a workflow guide, not a claim that every service has a dedicated Smalltown Boy preset or free generation allowance. Check the controls and cost displayed in your account.

Method 1: Higgsfield — use a motion reference

Higgsfield's Genjutsu overview distinguishes Motion Transfer from Object Swap: the first carries movement into a new scene; the second replaces selected elements. For a dance workflow, choose Motion Transfer and provide a reference you are allowed to use, together with clear subject images. See the official walkthrough for the current interface.

Describe the role assignment rather than asking for a different choreography:

Use reference image A for the performer who begins near the camera and image B for the performer behind. Preserve the reference movement, exchange timing and camera framing. Keep both identities consistent when they switch depth and finish together.

Inspect the exchange carefully before keeping a result: the two faces should not swap identities when their positions change. If that happens, simplify the cast descriptions and use more distinct references. No generation was purchased to write this guide.

Method 2: Kling — build the sequence from a starting image

Open Kling's creation workspace. For an image-to-video workflow, prepare a starting composition showing both subjects with room to move. Paste the main prompt above, then adapt it to the controls your selected model actually exposes. Do not assume a single-person animation mode can bind two separate character photos.

For a more manageable first draft, split the scene into three clips:

1. Foreground reaction: A reacts near the camera; B dances behind.
2. Role exchange: A steps back as B comes forward, without crossing through each other.
3. Shared dance: Both remain fully visible and take small coordinated steps.

Keep the same subject descriptions, setting and framing across the clips. If your selected Kling model offers motion-reference controls, use its instructions for that mode; the image-to-video prompt alone will not reproduce an exact reference dance. The Higgsfield model guide also describes Kling Motion Control as reference-driven character movement.

Method 3: CapCut — assemble and time the final video

CapCut's Smalltown Boy trend guide describes preparing the motion first and finishing the edit afterward. Import your recorded footage or generated clips, place them in reaction → exchange → shared-dance order, then trim pauses so the change of foreground subject is easy to follow.

Add captions only where they help the joke. Preview the edit without audio to catch jumps or duplicate subjects. Add a track you are permitted to use, or choose licensed audio in the destination platform. Check template and export requirements before choosing paid features. A CapCut template may arrange clips; do not assume it creates a new two-person AI performance from still photos.

Fix the common problems

• Faces exchange when positions change: give each subject a distinct visual description and simplify the movement.
• Hands or feet merge: increase the space between subjects and reduce crossing movements.
• The model skips the final dance: shorten the reaction section or generate the final shot separately.
• The video looks right but feels off-beat: adjust clip timing in the editor; naming a song in a visual prompt does not synchronize it.
• The result differs from the reference: text guidance is approximate. Use a supported reference workflow when exact timing matters.

Use the prepared generator

The homepage brings the sample, two-photo inputs and generation settings together. Open the Smalltown Boy AI Dance Generator, preview the dance, and check the displayed credit cost before generating. The step-by-step two-photo tutorial explains the site's own flow.

Workflow references checked November 2, 2026. These are independent services; features and availability can change. Smalltown Boy AI is not affiliated with Higgsfield, Kling or CapCut.