I have been offline for two weeks and got surprised by the reddid blackout. AI developes fast, so what is the best actual model to train LORA for realistic character design in SD1.5 ?

  • siegekeebs@lemmy.dbzer0.com
    link
    fedilink
    English
    arrow-up
    3
    ·
    edit-2
    1 year ago

    It depends what model you are planning to use to generate. My personal favorite for realistic portraits is henxmix realistic. I find it to be much better than realistic vision personally. If you train a lora in one model though, it won’t necessarily work perfectly in other models. I did a comparison of training the same lora with a bunch of different base models and the results were all similar though.

    I often hear people say to train only on the base model though, I haven’t personally found a benefit from that, but worth mentioning.

  • idle@lemmy.dbzer0.com
    link
    fedilink
    English
    arrow-up
    1
    ·
    1 year ago

    for 2.1 expect di.ffusion.ai in a week or so along with an automated trainer based on K;)

    try training with hugg repo: FFusion/di.FFUSION.ai-v2.1-768-BaSE-alpha

    Please note that SD 2.1 is different in prompts (AND) mainly handles negative examples. If you are not yet familiar with these features, I suggest sticking with models from version 1.5 for now 🤟 🥃