This workshop breaks down the Character Window workflow, which combines server-hosted Daydream StreamDiffusionTD with MediaPipe to create a pinch-pulldown window over a person’s face in real time.
Here’s how it works:
- Pinch-pulldown effect: Triggered by MediaPipe hand tracking, a window overlays the user’s face.
- Real-time face replacement: The window displays the output of StreamDiffusionTD, which swaps the user’s face for a prompted visual filter.
- Dynamic prompt system: Prompts are pulled from a ChatGPT-generated master list and rotate between three modes: Celebrities, Fantasy Characters, and Animals.
We’ll walk through each part of the stack, from hardware and server setup to prompt generation and mode switching, so you can see exactly how these components work together.
Experience level