The goal of this thesis is to design and implement a modular framework for real-time video and audio processing that will maintain high performance and flexibility through hardware acceleration. The system will be based on a graph-oriented architecture that enables the dynamic composition of media-processing pipelines with a focus on streaming. The framework shall be able to load RTP and UDP streams that contain both video and audio, as well as still images in common formats. The solution must be designed in a framerate-agnostic manner, allowing the output to operate independently of the input framerate. There should be support for multi-output scheduling, enabling simultaneous rendering and streaming to multiple destinations. Additionally, the system should include support for blending the inputs and for creating simple animations. The visual representation of the graph will not be part of the thesis. Instead the a JavaScript scripting engine will replace a traditional user interface and enable the user to construct and control the processing graph dynamically through asynchronous JavaScript APIs. In the theoretical part, the student will discuss the principles of real-time multimedia processing, including synchronization, buffering, and latency management.