>> Prompt - When providing prompts, get creative! Try your favorite artists, instruments like saxophone or violin, modifiers like arabic or jamaican, genres lik

>> Prompt - When providing prompts, get creative! Try your favorite artists, instruments like saxophone or violin, modifiers like arabic or jamaican, genres like jazz or rock, sounds like church bells or rain, or any combination. Many words that are not present in the training data still work because the text encoder can associate words with similar semantics. The closer a prompt is in spirit to the seed image And BPM, the better the results. For example, a prompt for a genre that is much faster BPM than the seed image will result in poor, generic audio. (1) Is there a corpse the keywords were collected from? (2) Is it possible to model the proximity of the image to keywords and sets of keywords?