CivArchive
    yiffyfluffybutts - yiffyfluffybutts-v0.61
    NSFW
    Preview 10692146
    Preview 10694147
    Preview 10692152
    Preview 10692147
    Preview 10692148
    Preview 10692150
    Preview 10692157
    Preview 10692159
    Preview 10692156
    Preview 10692619
    Preview 10693865
    Preview 10693866
    Preview 10858998
    Preview 10859002
    Preview 10859048
    Preview 10861260
    Preview 10860015
    Preview 10860322
    Preview 10860949

    Please read these instructions for the best results.

    This model is really sensitive to artist styles (that's why I made it), and sensitive in general.

    Don't give it too much to think about, give it multiple inputs for the same body part, or try to force a result with ((((((gimmie this)))))) or results will be poor.

    It's honestly a bit too sensitive to inputs, but works well for specific artists. As you can see from the samples I like doxy/onta :)

    Most importantly: I made this model to be good at replicating my favorite artist(s). IF YOU GENERATE A LOT OF IMAGES FROM ONE ARTIST, FIND THEIR PATREON AND SUBSCRIBE. DO IT. DO IT NOW.

    The artists are still the creators of the images you are generating it's just via AI as an intermediary. It's really unethical if you don't support an artist whose skills and hard work you're enjoying. Pay them.

    Back to settings, I strongly recommend using an e621 tags extension in automatic1111 or looking up images on e621 to see what the correct tags are (e.g., 'scut tail' instead of 'deer tail') as they will work much better, gelbooru tags may also work because of the merged model.

    also the model is really sensitive to the number of hits on e621, so if you have a tag with <1k results it's not going to be expressed as reliably as a common tag with 50k. Try to use tags with lots of images, and if you have to use a low image count tag increase the strength with (tag) or tag:1.2

    the tags extension in automatic1111 also shows you how many results there are for tags, you should use it

    the settings I prefer are Heun, 100 steps, cfg scale 8

    seems to work better without hires fix or cfg rescale, ymmv

    you probably don't need character or artist loras as long as they have enough images for that tag on e621

    badhandsv4 works better than v5 for me

    ubbp seems to be the only other negative embedding needed, try them all yourself

    adding 'simple background' in negative and 'detailed background' in positive can help if it won't make the bgs you want

    making:

    First, I made the fluffybutts model with: fluffyrock-576-704-832-960-1088-lion-low-lr-e257-terminal-snr-vpred-e11 + (Gape model - NAI model)

    and multiplier of 1

    then this was combined with yiffymix like this:

    yiffymix_v43 * (1-0.65) + fluffybutts * 0.65

    to make the finished model

    Description

    FAQ

    Checkpoint
    SD 1.5

    Details

    Downloads
    329
    Platform
    CivitAI
    Platform Status
    Available
    Created
    4/25/2024
    Updated
    4/22/2026
    Deleted
    -

    Files