Jen uses two categories of AI models: audio and language. Our Omnidirectional Audio Diffusion Models have received a Fairly Trained certification for our fully licensed training approach. You can read more about the Fairly Trained certification here. The Omnidirectional Audio Diffusion Models are the models that generate music. The Fairly Trained certification for these underscores our commitment to fair and responsible AI practices. We currently use the T5-FLAN language model, which is a third-party model that we use solely to convert user’s text prompts into numerical representations known as text embeddings. These embeddings capture the semantic meaning of the text without generating any new content. The model does not generate any music. We maintain strict boundaries between the text processing of FLAN-T5 and the music generation of Jen’s audio diffusion models.