• NekoKoneko@lemmy.world
    link
    fedilink
    English
    arrow-up
    2
    ·
    23 hours ago

    I hear that, but it really depends on the service and prompt (including services’ internal prompt that is hidden) and result, which are many times black boxes.

    I personally think artists & labels will have a tough time proving infringement for non-infringing outputs based purely on training data. But there’s really no way of being sure that the “generated” and “uncopyrightable” AI track that’s distilled from unlicensed source music is not itself infringing as a pure substantial similarity (or whatever your locality’s infringement legal test is) question.