daydreamlive/scope

What it solves

Scope provides a way to run, customize, and interact with generative AI video pipelines in real-time. It addresses the difficulty of managing low-latency video generation and the need for a composable system where different models and processing techniques can be combined.

How it works

It uses a composable pipeline architecture that supports autoregressive video diffusion models (such as StreamDiffusionV2 and LongLive) and configurable VAEs. The system allows users to combine video diffusion with other techniques like frame interpolation and depth mapping. It features an interactive UI with a timeline editor and supports real-time streaming via WebRTC, NDI, Spout, and Syphon for sharing video across networks or local applications.

Who it’s for

It is designed for creators and developers who need real-time, interactive generative video capabilities, as well as those looking to build custom AI video pipelines using plugins and LoRAs.

Highlights

• Real-time Pipelines: Supports low-latency async video processing and WebRTC streaming.
• Composable Architecture: Allows mixing video diffusion with depth mapping and frame interpolation.
• Extensibility: Supports plugins for new models and visual effects, and LoRAs for style customization.
• Broad Integration: Integrates with NDI, Spout, and Syphon for seamless video sharing.
• Interactive Control: Includes a UI with a timeline editor, text prompting, and parameter controls.