Rooftop Skyscraper Trampoline Jump Stunt

Night tracking shot. A woman in a white mini dress runs across a rooftop helipad and jumps off the edge. 0-5s: Running and jumping into the void. 5-8s: Falling several stories toward a glowing rainbow-rimmed trampoline. 8-12s: Hitting the trampoline and bouncing back up with a backflip. 12-15s: Landing back on the roof edge as friends cheer. Realistic city lights and motion blur.

モデル

Wan

ワークフロー

motion transfer

長さ

15s

アスペクト比

9:16

カメラ

tracking shot, high angle, dynamic follow

スタイル

cinematic realism, night photography

モーション転送版

参考動画と自分のキャラクター画像と一緒に使ってください。

Use the reference video for motion. Keep the exact body motion, timing, and camera; replace the performer with a woman in a white dress; setting is a night-time city rooftop and street-level trampoline.

ネガティブプロンプト

text, watermark, logo, blurry, distorted face, bad physics, low resolution

元の投稿

THE PROMPT ISN’T THE MOST VALUABLE PART OF A VIRAL AI VIDEO ANYMORE the motion is. look at this clip: girl runs across a rooftop → jumps several floors → hits a trampoline → backflips straight back onto the roof you understand the entire setup in the first second and now your brain has to see the landing normally you’d try to recreate something like this with a giant text prompt and pray the model understands the camera, timing and body movement but there’s a much smarter way now use the viral clip itself as the motion reference Wan2.2 Animate can take a driving video like this and transfer the movement + expressions onto a completely different character or use its character replacement mode and swap the person while keeping the original movement, scene and camera behavior there’s already a native ComfyUI workflow for it, and people have open sourced full "trend copy" pipelines around the same idea so the stack becomes: 1. find a format that already holds attention 2. use the original as driving motion 3. replace the subject / character 4. regenerate the creative around your own idea no keyframing the stunt no manually rebuilding the camera move no hoping a text prompt invents the same physics this is the part of AI video I think people are sleeping on viral videos are slowly becoming motion templates. the footage is disposable. the movement is reusable. workflows below 👇