Changing a scene with Ai

PERFORMANCE TO AI SCENE

Turn Your Living Room Into a Concert Stage

Film yourself at home, then use AI to put the same performance in a whole new world,

and show all three versions side by side in one vertical video.

What you need: a phone or camera on a tripod, a Higgsfield account (for the AI image and motion transfer), and CapCut (for the split-screen edit).

1 Record your performance

• Put the camera on a tripod. It must not move at all during the take.

• Shoot horizontal (16:9), in good light, with your full upper body and hands clearly in

frame.

• Hold a clear, strong pose in the first second. That first frame becomes your AI starting

image.

• Record a second, wider angle at the same time if you can. That is your

behind-the-scenes (BTS) clip.

2 Grab the first frame as a still

• In CapCut, move the playhead to 0:00 and export that frame as an image (or screenshot

it at full resolution).

• Use the frame from the start of the clip so the AI image and the video line up perfectly.

3 Transform the still into a new scene (Higgsfield)

• Upload the still to Higgsfield and pick an image-editing model (we used GPT Image 2.5,

quality High, 2K, 16:9).

• Start your prompt by locking the camera, face and pose, then list only what should

change. Template: Edit this photo. Keep the exact same camera position, angle, lens and framing, and keep my face, hair, body, pose and position in the frame identical.

[CHANGE 1: outfit, e.g. a tailored black tuxedo with bow tie]

[CHANGE 2: the instrument/prop, e.g. a glossy white concert grand piano, shot over

its closed lid, not blocking my hands or body]

[CHANGE 3: the object in my hand, e.g. a white dove perched on my hand]

[CHANGE 4: the location, e.g. a concert hall stage, red velvet curtains, warm

spotlight, haze]

Photorealistic, natural hands with five fingers.

4 Check the image before you animate

• Same camera angle and framing as your original? Your face still looks like you?

• Hands, fingers and body fully visible, with nothing new blocking them? Props look real?

• Generate one image at a time and fix the prompt based on what is wrong, instead of

re-rolling blindly.

PRO TIP

Think about how the new prop would really sit from your camera angle. A grand piano extends toward the

audience, so if the camera faces you from across the keys, describe it as seen over the closed lid.

Otherwise the AI will put your legs inside the piano or hide your hands.

5 Transfer your motion onto the new scene (Higgsfield)

• Open Genjutsu Motion Control in Higgsfield.

• Add the AI image as the image reference and your original performance video as the

video reference. Set 1080p.

• Prompt:

Transfer the exact body, arm, hand and finger motion from the reference video onto

the person in the image. Keep his face, outfit, props and background exactly as in

the image. Keep the camera static and matching the image framing. Photorealistic.6 Review and download

movement.

• Download the final render as an MP4.

• Watch the hands, the props and the face all the way through, especially fast finger

7 Build the 3-way split screen (CapCut)

• Start a new project with a 9:16 canvas.

• Stack three 16:9 clips top to bottom, each scaled to the full width: BTS on bottom, the

original reference shot in the middle, the AI result at the top.

• Line them up on the timeline so the motion is in sync across all three.

• Keep one audio track (usually the original performance) and mute the others.

• Optional: add small labels such as "BTS", "REAL" and "AI".

8 Export and post

• Export at 1080x1920 and post it to TikTok, Reels and Shorts.

• Hook idea for the first seconds: "I turned my living room into a concert hall."

Made by SPENCE