01 April 2019
I wrote some documentation and basic demo code for integrating OpenRNDR with RunwayML, a framework for finding, running, and working with machine-learning models for creative coding purposes. Shown above are the two demos - a facetracker and a pose estimator.
It uses SocketIO (although a feature of the individual models seems to be the ability to dictate the output format) for reading and displaying the output. I kept the display of the out as clean, simple and on-brand for OpenRNDR as possible.
RunwayML is an interesting idea - removing the developer skill requirements of finding, loading, and dealing with machine-learning models and creating a tool which is optimised for creative design and development using their output. I’m not sure what the market size is for the service but the pricing model is interesting, with the service charging for cloud-based GPU time (though it’s entirely possible to run the models on your local machine - either CPU or GPU-enhanced, depending on the model - at far lower speed). I’m also not sure what the motivation would be for model creators to share and tailor their models for the service.
Still - fun to play with.