Model: Midjourney
Prompt: a lemming drowning in baked beans while a crowd points and laughs
Effort level: Zero effort, got it in the first try


Not what I intended, but I couldn’t keep this to myself.
The prompt… well… just don’t ask :)

  • LongWinter@lemmynsfw.com
    link
    fedilink
    English
    arrow-up
    1
    ·
    1 year ago

    I tried… but couldn’t get anything as good as yours in stable diffusion. This is the best…

    It looks like midjourney is automatically adding context to the prompt? That crowd certainly looks good. Time for SDXL…

    • Thelsim@sh.itjust.worksOPM
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 year ago

      It’s funny because your image was exactly what I was trying to go for. It has:

      1. Actual beans and not some weird red mush
      2. A lemming that’s (kind of) drowning in it
      3. A crowd of which at least one of them is pointing and laughing.

      I kind of stumbled on mine by chance, but the other variations had the exact same mob. Midjourney definitely adds it’s own style to the mix.

      The prompt was originally meant as a commentary on everyone jumping on the bean bandwagon. It just turned out so weird and gruesome that I had to share it.

      Keep me posted if you can actually get a good composition, I’ll pass the torch on to you :)

      • LongWinter@lemmynsfw.com
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 year ago

        Thanks! You’re right about the prompt, your image doesn’t really match it but is much more appealing visually. My image above was made with SD2.1, I can get nicer looking stuff with 1.5 models, but then I have to be much more specific with the prompt. Hopefully SDXL will improve all of that.

        I made a batch of 10 and this is the only image that actually put a lemming into beans…

        • Thelsim@sh.itjust.worksOPM
          link
          fedilink
          English
          arrow-up
          0
          ·
          edit-2
          1 year ago

          I don’t have a lot of experience with SD, but I do notice you get nicer results when you mention a specific style or artist (Hi Greg!)
          Maybe if you combine it with the style of a naturalist artist like John James Audubon and another more generic style, you get more consistency?

          By the way, can I say that I love having this discussion. It’s nice to have a back-and-forth over these kinds of things instead of a single comment :)

          edited because I’m impatient and didn’t review my post

          • LongWinter@lemmynsfw.com
            link
            fedilink
            English
            arrow-up
            1
            ·
            1 year ago

            I did add an artist for the image above, Gerald Brom. That definitely makes it more interesting visually. The problem is that the 1.5 models I have seem to be trained with them, but they don’t really understand the prompt well, this is the best I got.

            The 2.1 model I used does a good interpretation of the prompt, but apparently wasn’t trained with different styles. It’s probably built for photo realism. So I tried a different 2.1 model and got this, which seems to go in the right direction.

            Lol, I should get back to work.