Diffusion As A Live Tool : Building Real-Time AI Visual Systems in TouchDesigner (Day 2)

Session details:

StreamDiffusionTD transforms a generative AI model into something TouchDesigner artists already understand: an operator in a network. Drop it into a patch, feed it a signal, and it begins producing unique, original visuals in real time—delivering up to 30 FPS of AI-generated imagery directly inside your workflow.

This workshop focuses on the practical realities of integrating that operator into a live performance pipeline: what inputs work best, how to control and process the output, and how to push the system for maximum performance. We'll begin with a brief overview of StreamDiffusionTD's architecture and underlying mechanics, then move into hands-on techniques for live and generative applications.

Topics include optimizing for real-time frame rates, effective use of noise and feedback, prompting strategies, leveraging ControlNets to guide imagery through structure rather than language, and post-processing workflows for refining results. If time and interest allow, we'll also explore audio-reactive approaches and the creative possibilities they unlock.