Real-time Stable Diffusion visuals inside TouchDesigner
This tool is experimental or a prototype. Don't use it on a live show without testing.
ExperimentalTouchDesignerAdvancedTouchDesignerStreamDiffusionNVIDIA GPU
What it does
TouchDiffusion brings StreamDiffusion into TD for interactive, low-latency generated imagery driven by live inputs.
How to set it up
- Easiest route: download the portable build linked in the README, run webui.bat and use the install & update tab.
- Manual route: install TouchDesigner 2023, Python 3.11, Git and CUDA Toolkit 11.8, then follow the README or its YouTube tutorial.
- Build an engine for your model (the README benchmarks sd-turbo at 512x512) and load the component in TD.
- Feed a live TOP (camera, LED content) in and route the output to your media server.
Source project: olegchomp/TouchDiffusion
FAQ
How fast is it?
README benchmarks: about 55–60 FPS on an RTX 4090 and 16 FPS on an RTX 3060 12GB at 512x512.
Is it stable?
The author marks it as early testing that may change significantly.
Original source: https://github.com/olegchomp/TouchDiffusion