squatters

Learning TouchDesigner for a project - possibilities, limitations, workarounds. Currently acts as the end point of the whole process, but if I can get familiar enough with it might become the main UI for this real time AI project. One thing I need is background removal so I started with that. There's a node for this but it requires a Nvidia card, so while that works fine on my desktop (screenshot here) it won't on my macbook, so I need to switch to using the 'Open Broadcaster Software (OBS) with background removal plugin' instead. While I see how far I can go with TouchDesigner, in the meantime the main user interface for the project is a web page. This does the webcam, a very basic background removal (although OBS could do that job here too I like the noise and so does the AI!), p5js for all sorts of dynamic background graphics (sunrise example here), and websockets for sending images, data, and receiving results images (as preview). Oh, and mock EEG values until I can work with real live EEG data, because that is to be the main feature! There's also a cloud based ComfyUI instance which is running a continuous real time AI process. Receiving data via websockets, and spitting out the result as frames via websockets. It's really fast, which is nice. Prompt text and some other settings are soon going to be remote controlled, so it'll just be a case of launching it and leaving it alone to do it's thing. The output frames from ComfyUI go to a websocket and the initial web interface picks them up for preview. TouchDesigner can also pick these up (with a little bit of Python coding), and further processing can be done along with other work for display/projection etc. I can imagine even more complex setups, but happy to have proven all these steps so far! At some point with this project I'm going to satisfy my curiosity about what happens when you feed/mix the AI output back into the same real time process... – at Edinburgh, Scotland