Wow, that's incredible! Any chance we'll get a downloadable and editable open source version of this so we can play around with it, train it on smaller own datasets and generally experiment? It sounds super exciting!