A prompt is not a magic spell. In a photo-to-dance workflow, it is a set of constraints: what to preserve, what to change, what to avoid. The scene supplies the choreography. The photo supplies the identity. The prompt is the contract between them.
Most prompt guides ignore that and give you a paragraph of mood words. The result is a clip that looks nothing like the photo and a wasted credit. This guide explains the four jobs a prompt can actually do, gives you 30 patterns, and shows you how to test them without burning money.
The four jobs of a dance prompt
1. Protect identity
The most important job. Tell the model what must stay: face, facial proportions, skin tone, hairstyle, glasses. Be specific about the features a human would use to recognise the person.
Example: “Preserve the uploaded person’s face, eye spacing, glasses and short dark hair. Keep the same number of people as the photo.”
2. Protect the outfit
If the outfit is the point, say so. Name the colours and the pieces. Tell the model not to invent new clothing or change the shoes.
Example: “Preserve the black jacket, silver zipper and white sneakers. Do not add jewellery or change the silhouette.”
3. Protect the frame
If the clip will be cropped to 9:16, ask for the head and hands inside the central safe area. If the feet matter, ask for a full-body frame. If the scene is wide, do not ask for a close-up.
Example: “Keep the full body and both shoes in frame. Do not crop the feet.”
4. Prevent the known failures
Name the failure you have seen: extra fingers, a second person, swapped identities, text, watermarks, a warped face. A negative constraint is often more useful than another positive adjective.
Example: “No extra people, no duplicated hands, no text, no logos, no watermark.”
30 prompt patterns
The patterns below are short on purpose. Use one at a time; do not combine them into a paragraph.
Identity patterns
- Face lock: “Preserve the face, facial proportions, skin tone and hairstyle exactly.”
- Eye lock: “Keep the same eye spacing and eye colour as the photo.”
- Glasses lock: “Keep the glasses on the face in every frame; do not warp the frames.”
- Hair lock: “Preserve the hair colour, length and parting.”
- Beard lock: “Preserve the beard shape and density.”
- Age lock: “Preserve the person’s apparent age; do not smooth or rejuvenate the face.”
- Expression lock: “Keep the neutral, confident expression from the photo.”
Outfit patterns
- Full outfit lock: “Preserve the outfit’s colour, fabric and visible accessories.”
- Shoe lock: “Preserve the shoes; do not replace them with generic sneakers.”
- Costume lock: “Preserve the costume’s silhouette and small colour accents.”
- Dress lock: “Preserve the dress silhouette, colour and visible details.”
- Uniform lock: “Preserve the uniform insignia and colour; do not invent new patches.”
- No new jewellery: “Do not add necklaces, rings or earrings that are not in the photo.”
Framing patterns
- Full body: “Keep the full body and both feet in frame.”
- Three-quarter: “Keep the head, arms and waist in frame; do not crop the elbows.”
- Vertical safe area: “Keep the head and hands inside the central vertical safe area for a 9:16 crop.”
- Wide scene: “Keep the wide composition and the space around the dancer.”
- No close-up: “Do not turn the clip into a close-up portrait.”
Motion patterns
- Sharp arms: “Follow the scene’s sharp arm shapes; do not soften them into flowing motion.”
- Footwork: “Follow the scene’s footwork and bounce; keep the feet visible.”
- Slow turn: “Keep the scene’s slow head turn and subtle expression.”
- No new dance: “Do not invent a new dance; follow the source scene’s choreography.”
- Same timing: “Keep the scene’s original timing and camera movement.”
Couple patterns
- Role mapping: “Use the first photo for the left performer and the second for the right.”
- No swap: “Do not swap the two identities at any point.”
- No third person: “Do not add a third person to the scene.”
- Matched style: “Keep both outfits’ colours consistent with their photos.”
Safety and output patterns
- No text: “No text, logos or watermarks anywhere in the frame.”
- No extra limbs: “Keep exactly two arms and two legs; do not duplicate hands or fingers.”
- Private result: “This result is private and should not be offered to the public gallery.”
How to test a prompt without wasting credits
- Pick one priority. Face, outfit, framing or failure prevention. Write one sentence for it.
- Use a scene you already know. Do not change the scene and the prompt at the same time.
- Generate once. Compare the result to the photo, not to a different clip.
- Change one variable. If the face drifted, keep the scene and change the identity sentence. If the outfit drifted, change the outfit sentence.
- Stop when it is good enough. A second clip that fixes one detail but breaks another is not progress.
The biggest waste is generating the same prompt five times and hoping for a different result. The model is deterministic enough that the same inputs usually produce a similar output. Change the input, not the number of attempts.
Why prompts cannot invent choreography
The tool edits a prepared performance. The dance comes from the source scene. A prompt that says “do the running man” will not add a running man to a scene that does not contain one. It can only preserve, constrain and style what is already there.
That is a feature. It is why the motion is coherent and the camera is stable. If you need a specific move, choose a scene that already contains it, or film yourself and use a motion-control workflow.
Common mistakes
- Too many goals in one prompt. The model follows the strongest constraint and ignores the rest.
- Contradicting the scene. Asking for a close-up in a wide scene, or a slow turn in a fast scene.
- Vague adjectives. “Cinematic” and “cool” do not tell the model what to preserve.
- No negative constraints. If you have seen extra fingers, say “no extra fingers”.
- Changing everything at once. You cannot learn what worked.
- Expecting a new dance. The scene is the choreography.
What to do next
Copy a prompt from the AI dance video prompt library, paste it into the direction field, and generate one clip. If you want the full workflow, read the AI dance generator guide. If you are making a couple clip, read how to make a couple dance video from two photos.
