Ramalama AI prompt (copy and paste costume reveal)
A copy-and-paste Ramalama AI prompt for the boots-first Halloween reveal, shot by shot, with costume and group versions and fixes for a face shown too early.
6 min read · Updated
The Ramalama prompt
Here is a complete Ramalama AI prompt for an image-to-video tool with one reference photo of your face. It keeps the camera at your boots for six seconds, then whips up to the reveal, which is the one thing that makes the trend work.
Ramalama witch reveal
Vertical 9:16, 10 seconds, photoreal, one continuous handheld phone take with no cuts, night, a suburban sidewalk lined with glowing jack-o’-lanterns and warm porch string lights, fallen leaves. Use my reference photo for the face. The face must not appear until 6 s. 0–6 s: the phone is a few inches off the ground, pointing up. Only black lace-up boots, legs and the hem of a long black velvet cloak stride toward the lens, one confident step per beat, the cloak swinging. 6–7 s: the camera whips up fast from the boots to the face, with motion blur. 7–8 s: the face from my photo under a tall pointed witch hat, a sly smile at the lens, a small shoulder shimmy. 8–9 s: a black-gloved hand touches the hat brim, framing the face like a photo. 9–10 s: a big laugh, a wave, a hand on the hip, a last little strut. Sound: no music, no voice; I will add the song. Avoid: showing the face before 6 s, cuts, a second person, text, a changing costume, a different face.
Rather skip the prompt? The Ramalama maker makes this reveal from one photo, with the song already in place.
How to use it
- Open an AI video tool with an image-to-video or reference-image mode, such as Kling, Higgsfield, Runway, Veo or CapCut’s AI video.
- Add one clear, front-on photo of your face.
- Paste the prompt, set vertical 9:16 and 10 seconds or the closest length.
- Generate, then check the first six seconds: no face allowed.
- Add the Ramalama (Bang Bang) sound when you post and line the whip-up with the whisper.
What each line does
These are the beats of a finished Ramalama reveal. Each timed line in the prompt asks for one of them.
- “A few inches off the ground, pointing up” is the camera. Without it, tools film you at eye level and there’s no reveal.
- “One step per beat” slows the walk down; AI walks tend to rush.
- “Must not appear until 6 s” is said twice on purpose, in the header and in the timing.
- The three poses at the end give the clip a finish instead of a frozen smile.
Other costumes, men and groups
Swap these lines in to change the costume or the cast. Keep the camera and timing lines as they are.
Vampire
0–6 s: only shiny black boots, dark trousers and the hem of a long black cape with a red lining stride toward the lens. 7–8 s: the face from my photo, hair slicked back, a high red-lined collar, a slow knowing smile.
Little devil horns
0–6 s: only black heeled boots and the hem of a short black cape stride toward the lens. 7–8 s: the face from my photo with small curved black horns, a sly smile, a little shimmy.
Your own costume
0–6 s: only [shoes] and [the bottom of the costume: a hem, a tail, a cape] stride toward the lens. 7–8 s: the face from my photo wearing [the costume top and headpiece], a sly smile at the lens.
Two or three friends
Use my reference photos, one face each. 0–6 s: only the friends’ boots and costumes walking side by side toward the low lens. 6–10 s: the camera tilts up and reveals each face one after another, left to right, then they all laugh together.
Use your own face, or a friend’s with their permission. Don’t upload a celebrity or anyone under 18.
Fix it: what to change when it goes wrong
| What you see | Change in the prompt |
|---|---|
| Your face shows in the first second | Move “the face must not appear until 6 s” to the very first line |
| The camera is at eye level | Write “phone lying almost on the ground, tilted up at the boots” |
| It cuts instead of tilting up | Add “one unbroken shot, the camera physically tilts up” |
| The walk is too fast | Write “slow, heavy steps, one per beat” |
| The costume changes after the reveal | Describe the same cloak, boots and gloves in both the 0–6 s and 7–8 s lines |
| The face drifts from your photo | Use a sharper, front-on photo and add “the same face as the reference photo” |
Photo tips
The whole trend is one second of face, so that second has to look like you. Use a sharp photo, face to the camera, eyes visible, in good light. A head-to-knees photo also helps the tool keep your build. More in the photo guide.
What a prompt can’t do
- It can’t add the song. AI tools don’t play Ramalama (Bang Bang); you add it when you post, as our sound guide shows.
- It can’t time the reveal to the whisper. You line it up afterward by moving the sound’s start point.
- It can’t promise the face stays hidden. Expect a few retries.
The no-prompt route is the Ramalama page: one photo, the witch costume, the reveal on the whisper and the song included. Packs and plans are on the pricing page.
Independent fan tool
NowStarring is an independent fan tool, not linked to Róisín Murphy or the makers of Ramalama (Bang Bang). The song belongs to its rights holders.
Questions
What is the prompt for the Ramalama AI trend?
A vertical, 10-second, one-take video: a camera near the ground follows only your boots and cloak for six seconds, then whips up to your face for a smile, a hat-brim pose and a laugh. The full version is at the top of this guide.
Can I use the Ramalama prompt in Higgsfield or Kling?
Yes. Use their image-to-video or reference-image mode with one face photo, then paste the prompt.
How do I change the costume in the Ramalama prompt?
Use the “your own costume” lines: describe the shoes and hem for the first six seconds and the top and headpiece for the reveal.
Why does my face show too early?
Tools like to show the face. Put “the face must not appear until 6 s” first and describe the camera as almost on the ground.
Does the Ramalama prompt include the song?
No. Add the Ramalama (Bang Bang) sound when you post.
Can men do the Ramalama trend?
Yes. Use the vampire lines or describe your own costume; the walk and the reveal work for anyone.
How long should a Ramalama video be?
About 10 seconds: six of boots, then about four of poses after the reveal. Longer clips tend to lose the joke.
