cross-posted from: https://lemmy.today/post/59448913

Created by me using my own locally run Stable Diffusion XL 1.0 (SDXL) model with a custom LoRA trained with my own acrylic paintings and ink washes, with additional editing in GIMP and Inkscape.

Based on my own sketches. Original sketch in comments.

“I tried doing them by hand, but I find it easier to use a screen. Don’t think about making art, just get it done." — Andy Warhol, 1963

  • PapaSkwat@lemmy.todayOPMEnglish
    1·
    2 months ago

    Here’s the photo of original sketch and layout that was basis of this. It was a concept I thought of a bit ago, but I never got around to updating and actually painting. Painting is expensive so I go through a lot of concepts before I actually put something on canvas.

    So this never made it to canvas, but it’s made it to AI! All hail our Lord and Savior! May my Slop be Forever in the Cloud!

    My original drawing that AI used for reference:

      • PapaSkwat@lemmy.todayOPMEnglish
        1·
        2 months ago

        Thank you! Since I feed my own artwork into my AI as reference, and I’m detailed with my prompts, it usually gets pretty close to what I want. Every once in a while it’ill totally go off-script. but for the most part, it does what I want.

        I’m a painter in real life, so I usually specify the palette in the prompt. Sometimes I’ll just give it one of my paintings as a color reference and tell it to match. I mix and use a blue/green/red wash quite a bit, but for whatever reason the AI has a hard time reproducing it and tends to muddy everything toward brown.

        I even painted an old used canvas with just that wash, photographed it, and gave the photo to the AI as a reference. It still can’t quite get it. Could be because I photograph my physical stuff with a crappy $150 phone, as you can tell from the photo of my sketch. Other times I just let the palette go random.

        All of the models I use are uncensored.

        My main use for AI is composition. Sometimes I get bored with my original layout, so I’ll let the AI reinterpret the composition several diff ways. Sometimes I go back and adjust my physical canvas painting based on one of them.

        The AI is much better than I am at quickly exploring different compositions. I’ve ruined plenty of my own paintings by going into them with only a vague idea of the final composition, then realizing after I’ve painted half the thing that it’s just not working out.

        People who cry “slop!” at my stuff, aren’t usually artists, otherwise they’d know that AI is just another tool. But it’s Lemmy, so nuance isn’t really a thing on here. lol