STEP 1 - Record your performance shots
Film your performance scenes like you normally would. My biggest tip: make sure your lighting is solid so the AI can clearly see your movement and facial details.
STEP 2 - Build a rough edit
Import your footage into your editor and start laying out the timing and pacing you have in mind. This will serve as the blueprint for your final video.
STEP 3 - Export still frames
Capture still frames from the beginning of each clip. These screenshots will become the starting point for your AI-generated scenes.
STEP 4 - Generate your cinematic scenes
Upload each still frame individually into the AI platform of your choice, along with a reference image. I typically use Nano Banana Pro or Nano Banana 2, and I often include reference images from Pinterest (or other sources) to help guide the AI toward the exact look I'm after.
Quick tip: I upload my reference images into ChatGPT first and have it generate a detailed prompt for Nano Banana. This makes it much easier to maintain character consistency, framing, and the overall visual style across every image.
AI Scene Generation Prompt
Reusable Prompt
Replace only the environment while keeping the subject completely unchanged.
Preserve the subject's exact identity, facial features, expression, hairstyle, skin tone, clothing, body proportions, hand placement, camera angle, framing, crop, and perspective. Do not regenerate or reinterpret the subject in any way. Place the subject in [DESCRIBE YOUR SCENE].
The subject should naturally interact with the environment while maintaining the exact pose and expression from the reference image.
The camera should remain in the same position unless otherwise specified.
Lighting should match the environment naturally and enhance the mood of the scene.
Color grade the image with a cinematic look that fits the environment.
Prioritize realism while preserving the subject exactly as shown in the reference image.
The final image should feel like a frame from a high-budget film.
STEP 5 - Animate the scenes
Once you're happy with the generated images, use your original performance footage along with the AI images to recreate the motion and turn them into video clips. I typically use Seed Dance 2.0 for this step.
Quick tip: I also use ChatGPT to help write prompts before generating each clip. It does a great job describing movement, camera angles, and cinematic action, resulting in smoother and more consistent animations.
AI Motion Transfer Prompt
Reusable Prompt
Use the reference video as the source of all movement.
Copy the subject's body movement exactly, including head movement, shoulder movement, arm movement, hand placement, posture, and timing.
Use the reference image for the subject's identity, clothing, environment, and composition.
Animate the scene while preserving:
• Facial identity
• Clothing
• Camera angle
• Framing
• Perspective
Allow the environment to move naturally while keeping the camera locked.
Add realistic environmental motion that matches the scene (wind, rain, dust, light streaks, reflections, etc.).
Keep the subject synchronized with the original performance while maintaining a cinematic, realistic appearance.
Constraints
Do not change the subject's identity.
Do not alter clothing.
Do not change the camera angle.
Do not add new actions or gestures.
Do not zoom or pan the camera.
Keep all movement faithful to the original performance.
STEP 6 - Bring it all back into your editor
Import the generated clips into your timeline, blend them with your original footage, color grade, add effects, and make the edit your own.
Most importantly… experiment and have fun with it