Lock the camera
Use a tripod or stable surface. Avoid zooming, panning, and reframing.

Record a short hand-frame clip, generate a pixel-aligned 3D, anime, cyberpunk, or watercolor layer, then use four-fingertip tracking to reveal it only inside the moving opening.
Build your effect
Reliable finger-frame workflows restyle the full source clip first, then use fingertip tracking for the final moving mask. These presets change the look while locking timing, camera, face, and motion.
Ready-to-run prompt
3D Movie Portal
Output frame
Opening the workspace does not start generation. Use your own 4–10 second MP4/MOV and review the credit cost. After export, combine the source and styled layer in the GitHub tracker.
Record for stability
The model only creates the aligned style layer; the browser tracker creates the moving finger mask. These four recording choices make both stages more reliable.
Use a tripod or stable surface. Avoid zooming, panning, and reframing.
Keep both hands visible and hold a clean finger window for at least two seconds.
Use soft front light so every fingertip stays separate from the background.
Generate the shortest layer first. Check frame alignment, hands, face, and camera before spending credits on a longer clip.
Authorized trend references
These credited X clips document the visual language behind the trend. They are references, not outputs from C Dance AI. Use your own footage when generating.
A video model can restyle a clip, but it does not reliably hold a moving four-point mask by itself, and even an alignment prompt can drift. Generate the four-second layer first and inspect it. The GitHub tracker then clips the accepted layer inside the four fingertips and keeps the source outside.
Open source and hosted
C Dance AI creates the aligned 3D, anime, cyberpunk, or watercolor layer. Download it, then load the original clip as Source video and the generated layer as Portal video in the GitHub project for local fingertip tracking and export.
It combines two videos with a four-fingertip mask: the original stays outside the hand-made frame, while a pixel-aligned AI style layer appears inside it.
Use a short, well-lit clip with a locked camera. Keep both hands visible and hold the opening for at least two seconds.
Gemini Omni video edit is preselected to create the aligned style layer from your source video. Fingertip tracking and the final mask happen separately in the browser.
No. The dependable workflow uses C Dance AI for the aligned style layer and the GitHub tracker for the moving four-point mask. This avoids asking the model to guess the mask on every frame.
Only use footage you own or have permission to edit. A public post does not automatically grant remix or redistribution rights.