• 🍉 Albert 🍉@lemmy.world
    link
    fedilink
    English
    arrow-up
    6
    arrow-down
    1
    ·
    1 day ago

    instead of writing pages of dialogue, write a lot of back story, personality, interests, knowledge, info they have, quests they have to share, sample of how they talk…

    fine tune models… this way each character would sound unique, rather than standard chat gpt.

    a good prototype would be about a village with about a dozen of NPCs.

      • JackbyDev@programming.dev
        link
        fedilink
        English
        arrow-up
        1
        ·
        13 hours ago

        I feel you, but imagine if a game dev trained their own LLM exclusively on their own content. Only their own dialogue and lore. Could be fascinating then. Not something I wanna see a lot of work out into at the expense of other things, but could be interesting.

        And to reiterate, I genuinely mean only trained on data the studio has a copyright on already.

      • 🍉 Albert 🍉@lemmy.world
        link
        fedilink
        English
        arrow-up
        6
        arrow-down
        1
        ·
        1 day ago

        another use, draw assets for a age of empires like game. then generate a diffusion model on them. now you can make rows of houses and non of them will be identical and all will fit in the art style.

        same things with textures, no more repeating textures.