So, the main goal may sound ambitious, but I really want to push the limits of real-time Stream Diffusion generation and find out, if it's possible to use it in production for real-time meaningful diffusion of video input.

CONSTAINS:

- 2-3 diffusion steps

- ControlNet for object boundaries coherence

- V2V for frame coherence

- Kinect azure / Orbbec input for real-time masking using point cloud data

- Local inference on NVIDIA RTX 4090

- TouchDesigner integration with interactive parameters control.