Gacha splash LORA

    GachaSplash4.safetensors LORA / SD 1.5

    Model Information
    Model Name
    Gacha splash LORA
    Version
    v4.0
    Size
    144.13 MB
    Downloads
    49,061
    Trigger
    (white background:1.5)[(white background:1.5),::5]
    Torrent Details
    BTIH
    9432B73F2B87804B9D1752665C3C5E7A1C3AB91E
    BTMH
    D08B69ABFA6CBC75E8A24DCA3F016F3A272FC8B5149506298C8DD61E601EA7CB
    SHA256
    EEBDB8E46FAFDAF3FCA98D0B3ADB0CBB16F71E138B65DDF8E43288D0B4A511EB
    Upload Date
    about a year ago
    Uploader
    CivitasBay.org
    Status
    2 Seeders
    0 Peers
    Info

    The training dataset is 100% generated by NovelAI. No real artwork is used.

    Checkout my Alter version, which uses real artwork: https://civarchive.com/models/85859/gacha-splash-alter


    3.0 version is released. I rebuilt the entire training set for this version.

    Notes:

    1. Use hires fix + SD upscale or MultiDiffusion to get better face details.

    2. Try adding wood, stone in negative prompt if you see too many woods, stones, structures. You may also emphasize it. For example, the bottle keeps getting transformed into structures till I add wood, stone.

    A 2.0 version is released:

    A gacha splash style lora. Make sure you start with the following template and add your background prompts.

    [(white background:1.5)::5], isometric OR hexagon , 1 girl, mid shot, full body, <add your background prompts here>

    Suggested resolution: 640X640 with hires fix. You may modify your prompt and resolution after successfully getting the effect.

    I trained two different versions with different learning rate and scheduler.

    Gacha Splash Fantasy fits more in abstract backgrounds.

    Gacha Splash is intentionally trained to be slightly overfit. It fits greatly for architectures.

    See compares from sample images

    Notes:

    1. If your characters are always wearing jackets/half off jackets, try adding "off shoulder" in negative prompt. You may further add "jackets"/ "bare shoulders" if the issue persists.

    2. Similar as the jacket issue, if you find any elements that are always being generated, try adding them into negative prompt first. The training dataset is well tagged. Therefore AI should be able to remove these elements.

    Prompts like hexagon and isometric help to shape the gacha splash effect. There are some alternative choices.

    Gallery
    This model contains NSFW content. Click to show gallery.
    Trackers
    • udp://tracker.opentrackr.org:1337/announce
    • udp://open.demonii.com:1337/announce
    • udp://open.stealth.si:80/announce
    • udp://tracker.torrent.eu.org:451/announce
    • udp://tracker.qu.ax:6969/announce
    • udp://tracker.bittor.pw:1337/announce