Hey everyone,
Today I'm launching something I've been building for a while: StreamDiffusion TOP, a real-time AI image generation plugin for TouchDesigner.
You can find it here : https://unveil-studio.lemonsqueezy.com/ and cost only 15euro (+ some taxes) for a 6 month license :-)
The idea is simple — diffusion models are usually something you render and wait for. I wanted to turn that into an instrument you can actually play live, inside your TD network, like any other operator.
So the controls are built around that. You get two prompts (Prompt and Prompt B) that you can blend between in real time — hook that up to audio or MIDI and you can morph the look of the image as it plays. Strength decides how much the AI reinterprets what you feed it versus staying close to the source. CFG Scale is how strictly it follows your prompt versus wandering off into something looser and more painterly. And Dry/Wet lets you blend the AI result back with your original input, so instead of an on/off effect you get a dial.
On the performance side, I didn't want to compromise: it runs at 75 fps at 512×512 with stream diffusion turbo, with optional 4X upscaling that only adds about 2.6 milliseconds per frame. All together it's roughly 3x faster than the equivalent PyTorch pipeline, and it runs on NVIDIA Ampere, Ada and Blackwell GPUs — so RTX 30, 40 and 50 series are all covered.
Before you dive in, check the manual that ships with the plugin — it walks through every parameter and gets you up and running fast.
That's the launch! Thanks for sticking with the project through all the development posts — this is the moment it actually turns into something you can use. More to come soon.
Best,
Simo








