MediaPipe — what creators actually do with it

Every MediaPipe post here is also a TouchDesigner post: TouchDesigner appears in all 6 of them. Creators are not calling the library from code; they use @blankensmithing's MediaPipe plugin inside TouchDesigner, which @neles0n tests on camera and a commenter calls "just built different" (plugin test). The main use is hand and body tracking as a live controller. In @pepepepebrick's spider piece, hand positions move two creatures across a generated 3D web, and leg count and distance feed pitch and waveform in VCV Rack, so one tracking feed drives both image and sound (spider web synth).

Other entries extend that tracking. A compiled guide pairs MediaPipe body tracking for interactive dance art with 3D scanning and Gaussian splats (creative coding guide), and @gerrylistudio's R&D piece, reposted by @touchdesigner, combines MediaPipe with 3D Gaussian Splatting and is labelled not a finished work (presence study). At DATLAB, @blankensmithing and @dotsimulate show free components that train classifiers on CHOP data such as gestures, image embeddings or EEG signals; they compare the idea to Google's Teachable Machine and say the classifier itself is a search algorithm, not machine learning (Classifier TOX).

The evidence is thin: six posts, several of them silent, and none covering setup problems, latency or tracking accuracy. @ojrgb's caption adds a note of caution about AI-assisted shortcuts flattening what is a deep craft (mixed feelings). The step-by-step version lives in the MediaPipe to TouchDesigner pipeline. It suits creative coders and installation artists who want camera-based hand and body input inside a real-time patch.