The exact prompt behind our renders, plus a pet version and a no-clip fallback for tools that cannot take a reference video. Copy any block into the video model you already use: the wording is what keeps the reference clip's motion, the camera and your photo's identity in place.
The complete prompt in one block. It opens with the identity anchor, describes the scene and the action beats in words, then the camera and the exclusions. The reference video rides along as the motion reference, so the prompt stays a description of what to generate - not a list of edits.
Create a ten-second vertical street-freestyle video. The performer is the person in the reference image: preserve their face, hairstyle, hair colour, skin tone, body proportions, outfit and accessories, and keep their identity stable for the whole clip. The scene is the entrance of a night-time venue washed in red light - red LED strips along the staircase steps, deep red shadows on the walls - with glass doors and a tight crowd of onlookers following alongside, most of them filming with phones raised. A friend walks with the performer holding a black round portable speaker up at chest height, so the speaker and the friend's shoulder keep crossing the foreground of the frame. The performance follows the action order and pacing of the motion reference: walking forward through the entrance while rapping straight at the lens, hand and arm gestures to the beat, raising a fist, hand to the chest, pointing past the camera, then a confident nod to the crowd. The camera stays close and handheld, walking with the performer and drifting around to their side the way a phone camera follows in a crowd. No duplicate subject, no new scenes, no face morphing, no extra hands, no extra fingers, no floating props, no style changes, no text, no watermark.Our generator applies this wording for you automatically - you only choose the photo. Copy it when you are driving another tool that accepts a reference video and want the same recognisable sequence.
Six moves, in the order the model reads them. Change one line and you change exactly one thing about the output.
The first two sentences name the performer and list what must not change - face, hair, clothing, accessories - and ask for that identity to hold for the whole clip. Keep it at the very top: it is the one instruction the rest of the prompt must never bury.
The venue, the red lighting, the staircase, the crowd and the speaker are written out in words - including the round speaker held at chest height and the friend's shoulder crossing the foreground. Describing them (instead of saying 'keep the reference') is what keeps the task a fresh generation rather than an edit.
The clip's beats - walking forward while rapping, the hand and arm gestures, the fist raise, the hand to the chest, the point past the camera - are listed in order, with the pacing taken from the motion reference. Writing the beats out is what keeps them recognisable.
The camera stays close and handheld, walking with the performer and drifting around to their side. It is described in words, and the reference video supplies the actual movement - which is why this line stays short.
Naming where the clip lands (the confident nod to the crowd) helps the model land the last second cleanly instead of drifting off.
The negative list is longer in a recast: no duplicate subject, no new scenes, no face morphing, no switching hands, no extra fingers, no floating props, no outfit flicker, no background identity swaps, no style changes, no text or watermark. These are the failure modes that actually show up.
Change the subject or the tool you are using, and keep everything else identical. Only the identity block (or the fallback) moves - the motion reference, the action order and the exclusions stay word for word.
Swaps the identity lines for fur markings, breed, face shape, ear position and proportions, and adapts the beats to small head and paw movement. The reference clip stays the same.
Create a ten-second vertical street-freestyle video. The performer is the pet in the reference image: preserve its fur markings, breed, face shape, ear position and body proportions, and keep them stable for the whole clip. The scene is a night-time venue entrance lit in red, with a red-lit staircase and a tight crowd of onlookers filming on their phones; the pet performs on the pavement beside a black round portable speaker held at chest height. The performance follows the action order and pacing of the motion reference, adapted to the animal: small rhythmic head bobs and paw lifts instead of rap gestures, a head turn to the crowd, then a cheerful finish. Handheld vertical close camera. No duplicate subject, no new scenes, no extra animals, no fur flicker, no style changes, no text, no watermark.Some tools cannot take a reference video. This version describes the street, the gestures and the camera in words instead, so you can still generate a video that reads as the trend.
Create the viral Bulin 47 street freestyle video from the person in the reference photo. The performer is exactly the person in the reference photo and no one else. Preserve their face, hair, clothing and accessories. Place them at night at the entrance of a venue lit in red, with a red-lit staircase, glass doors and a tight crowd of onlookers filming on their phones; a friend walks beside them holding a black round portable speaker at chest height. Use the recognizable sequence: walk forward while rapping straight at the lens, gesture with both hands to the beat, raise a fist, hand to the chest, point past the camera, glance at the friend's speaker, then finish with a confident nod to the crowd. Keep the camera handheld, vertical and close, circling the performer the way a phone camera follows in a crowd. Keep the motion natural, energetic and cinematic. Do not add text, logos, watermarks or extra faces replacing the main subject.Almost every failure traces back to one of four causes. Check them in this order before spending another credit.
Usually the photo: a distant or soft shot leaves the model very little to carry. Retry with a chest-up portrait in even light before touching the prompt.
Your prompt read as an edit instruction ('use the reference video as the exact motion and camera reference', 'match the reference'), so the model only rewrote part of the clip and left the rest of the source footage in place. Describe the scene and the beats in words instead - that is what the base prompt on this page does.
The identity line got buried. Put it in its own sentence right after the motion-reference sentence, and name the role explicitly - 'the lead at the centre of frame' - so the model knows which performer to replace.
Props multiply when the prompt names several at once. Keep one speaker and one gesture per beat, and keep the negative list short enough to be read.
If you would rather not write or tune anything, our generator applies this prompt to your photo, keeps the scene consistent between renders, and returns the credit whenever a render fails.
Reading and adapting the prompt examples is free. Each render uses one video from your pack, which covers the provider's processing cost.
Open the generatorNo. The prompt is fixed and tuned for this effect, so every render follows the same recognizable freestyle sequence. You choose the photo, not the wording.
Open prompts are the fastest way to lose the street scene: the model starts redesigning the scene instead of preserving the subject. Keeping the prompt fixed is what preserves the street, the speaker, the lighting and the camera move.
Yes - through the photo. Framing, expression, clothing, and visible features carry into the render. One clear subject facing the camera gives the most predictable result.