The Bulin 47 AI prompt, and when you need none
What you type depends on the app. A photo-to-video app needs a scene prompt: one performer rapping into a phone at night, a partner with a portable speaker, a small crowd, a red-lit doorway and a handheld vertical camera. A motion-control app that takes the real clip only needs one line saying who replaces whom. No prompt reproduces Kevin Wai's real footage on its own. On Bulin47Video.io there is no prompt at all: one photo goes into the real clip, with the original sound.
Updated
All postsTry it with your photo
Pick one photo and take Bulin 47's spot in the freestyle. The first 2 s are free.
Nothing is uploaded yet. You confirm it in the studio before anything is made.
Prompt or reference clip: what your app needs
Most "Bulin 47 AI prompt" videos skip this part, and it is why so many results look like a different street. A prompt describes a scene for the model to invent. A reference clip shows it the real one. Which you need depends on the kind of app, not on the trend.
| Kind of app | What you give it | What the prompt does | What you get |
|---|---|---|---|
| Photo to video (Dreamina, Kling, Veo and similar) | One photo and a text prompt | Describes the whole scene | A new, similar scene, not the Madrid footage |
| Motion control (Higgsfield Genjutsu, Kling motion control and similar) | The clip, your photo and a short prompt | Says who replaces whom | The same shot redrawn with you |
| One-photo template (Bulin47Video and others) | One photo | Nothing: the scene is already set | Depends on the site: the real clip or its own scene |
A scene prompt for photo-to-video apps
We wrote this one from the clip itself; it names no real person, which some apps refuse. Paste it, then change the bracketed parts to match your photo:
"Vertical 9:16 night video, filmed on a handheld phone at chest height. Outside a restaurant whose doorway glows red, [the person in the photo, wearing what they wear in the photo] raps an improvised verse straight into the camera. A young man beside them holds a small portable speaker up close and nods to a dembow beat. A loose crowd behind them films on their phones. The performer leans in and out of the frame, rolls their shoulders on the beat, points at the lens and laughs between lines. Natural street light, slight camera shake, real skin texture."
- Length: ask for 5 to 10 seconds. Longer runs drift away from your face.
- Photo: one person, facing the camera, shoulders in frame, no sunglasses.
- Sound: the app will not add the freestyle audio. Add the sound when you post.
- Dreamina's own Bulin 47 guide suggests a similar mix of elements and calls the result a recreation, not the real clip.
A swap prompt for motion-control apps
When the app takes the real clip as a reference, a long scene prompt only gets in the way: the camera, the timing and the crowd come from the clip. Say who changes and what must stay:
"Replace the man in the pink shirt with the person in the photo. Keep their face, hair and clothes from the photo in every frame. Keep the camera, the timing, the man with the speaker and the crowd exactly as in the video."
- Trim the clip to the seconds you want before you upload. Many apps cap the reference at 10 to 30 seconds.
- Use a full-body or waist-up photo. With only a face, the model invents the outfit and often keeps the pink shirt.
- Only use a clip you are allowed to use. Kevin Wai's post belongs to him.
Words that spoil the result
- "Bulin 47" or "Kevin Wai": many apps block or blur prompts that name a real person, and the ones that accept them may draw a lookalike instead of you.
- "Dance": the clip is a man rapping from the waist up. "Dance" gets you full-body moves and a wider shot.
- "Lip-sync to Dame Hombro": video models do not hear a song you name, so the mouth will not match it.
- "Crowd cheering, many people": more faces means more melted faces. "A loose crowd behind them" is enough.
- "4K, ultra realistic, masterpiece": filler that changes little. Spend the words on the scene.
When the result looks off
| What you see | What to change |
|---|---|
| The face drifts away from yours | A sharper, front-facing photo; shorter length; add "keep the face from the photo in every frame" |
| The pink shirt stays | Describe your own clothes in the prompt, or use a photo that shows them |
| Extra fingers or warped hands | Ask for "hands mostly at chest height", or pick a take where the hands move less |
| A daytime street | Put "night" and "red-lit doorway" at the start of the prompt |
| A music video look | Add "filmed on a handheld phone, slight shake, natural street light" |
Skip the prompt: one photo into the real clip
If you want Kevin Wai's real freestyle rather than a lookalike, a prompt cannot get you there. Our Bulin 47 video maker does that one job: you add one photo, and you get 15 fixed seconds of the original clip with you as the performer, the speaker, the crowd and the camera untouched, and the original sound laid back on. It takes about 4 to 15 minutes and costs 10 credits at 480p (the $9.99 Single pack). Each Google account can first make a free 2-second preview.
Other tools compared: Higgsfield, Dreamina and Genematic. The methods behind them: how the Bulin 47 AI trend works.
The speaker is already playing
Make my recastFAQ
Questions people ask
Is there an official Bulin 47 AI prompt?
No. Every prompt you see, including ours, was written by fans or app makers. Bulin 47 and Kevin Wai have not published one.
Can ChatGPT or Gemini make the Bulin 47 video from a prompt?
Their image tools make a still picture. For a moving video you need a video model, and even then a prompt makes a similar scene, not the real clip.
Can a prompt add the original sound?
No. Video apps either make no sound or make their own. Add the freestyle sound when you post, or use a tool that lays the original audio back on.
Do I need a prompt on Bulin47Video?
No. You upload one photo; the scene, the timing and the sound are already set.