Prompts & references
Smalltown Boy AI Video Prompt: People, Cats and Characters
A useful Smalltown Boy AI video prompt names two distinct subjects, assigns their roles and describes the reaction, exchange and joint dance in order. Exact choreography usually needs a suitable preset or motion reference. The examples below are editable starting points, not tested guarantees of identical output.
Choose the inputs your video model supports
Before pasting a prompt, check whether the selected model accepts two identity images and a video reference together. If it only accepts one starting frame, two separate uploads may not act as two persistent character references. Follow that model’s documented input format.
Use the original photos as identity inputs. Use an authorized dance clip as a motion reference where supported. Keep the written prompt focused on subject identity, role order, continuity and framing. Adding more adjectives cannot replace missing motion-reference support.
A two-person starting prompt
Assign Subject A and Subject B to the two identity images before using this example.
Create a vertical two-person dance sequence using the supplied identity images. Subject A begins in the foreground with an exaggerated open-mouth reaction while Subject B dances behind. They exchange foreground and background roles, then perform a synchronized dance side by side and finish facing the camera. Preserve each subject’s face, clothing and body proportions throughout. If a motion reference is attached and supported, follow its shot order and timing. Keep both identities distinct; do not replace them with the reference performers.
A cat-and-dog prompt
Name the animals’ distinguishing markings. Keep the species and coat details tied to one identity.
Use the supplied cat as Subject A and the supplied dog as Subject B in a playful vertical duet. Begin with the cat reacting close to the camera while the dog moves behind. Exchange their foreground roles, then show them dancing together. Retain the cat’s coat pattern, ears and muzzle, and the dog’s breed features and collar in every shot. Keep the two animals separate, with no merged bodies or species changes. Follow the attached motion reference only if the selected model supports it.
An original-character prompt
Use references for characters you own or are allowed to use. State which features must remain stable.
Animate the two supplied original characters in a vertical reaction-and-dance duet. Character A takes the first foreground reaction while Character B dances behind, then their positions exchange before a joint dance. Preserve each character’s silhouette, colors, outfit and distinctive facial features across close-ups and wide shots. Keep the same two characters through the final frame. Use the supported motion reference for the sequence, without importing its original performers or changing character identities.
What to change when the prompt misses the effect
| Problem | First thing to inspect |
|---|---|
| The dance is unrelated | Does the workflow have a preset or supported motion reference? |
| Faces switch after the exchange | Are the identities labeled consistently, with clear reference images? |
| Original performers remain | Did you select an overlay project instead of subject generation or replacement? |
| Pets merge or change species | Reduce scene complexity and keep one named identity per reference. |
| Beat or cut timing is wrong | Compare reference timing and the actual audio version in an editor. |
These are diagnostic steps, not a guarantee that a particular model will satisfy the sequence. Keep unsuccessful attempts and change one input at a time.
Template route or prompt route?
A fixed template may not need a written prompt. The no-prompt guide explains that alternative. Public reference workflows are described by Media.io and PicLumen. The examples on this page are original and have not been rendered or benchmarked by us.