• Funkytom467@lemmy.world
        link
        fedilink
        arrow-up
        18
        ·
        edit-2
        5 months ago

        Just checked and it is indeed an AI image based on a Asuka reference, with some change in Photoshop to make it seems less AI…

        It was posted a year ago on Baldur’s Gates subreddit. Technically no AI on the sub hence trying to hide it.

        https://www.reddit.com/r/BaldursGate3/comments/162xar0/command_grovel/?rdt=36411

        I found a pretty neat description of a process by the redditor GenderJuicy and generally arguing for it the coments :

        "Looks like he took this image and plugged it into ControlNet with a lower percentage, https://i.imgur.com/y3p7dN1.jpg, then used this LoRA https://civitai.com/models/106337/shadowheart-baldurs-gate-3-character-lora then took the output and used the Facet filter in Photoshop a few times to hide the “AI-ness”, thus the weird design inaccuracies, nebulous inhuman designs and such.

        Edit: Here you go, I did this in 2 seconds after downloading these and opening up Stable Diffusion, then used the Facet filter in Photoshop after pasting in the “Pathetic” text over it. https://imgur.com/0srdUM1

        If you’re going to make AI images and post them, don’t act like you “made it in Photoshop”. Pathetic."

        Also found a guy explaining it in the coments of a repost :

        https://www.deviantart.com/bitspirit3/art/Command-Grovel-979292113

        I gotta admit, the brush strokes got me too it’s pretty well done!

      • hswolf@lemmy.world
        link
        fedilink
        arrow-up
        9
        arrow-down
        1
        ·
        5 months ago

        this doesn’t give me the ai feel, the proportions are deliberate and there no meaningless mirroring of features, no expert either but this feels real

        • SuddenDownpour@sh.itjust.works
          link
          fedilink
          arrow-up
          4
          ·
          5 months ago

          I didn’t say “the AI would likely not draw clavicles”, but “the AI would likely not draw clavicles with a portion of which was remarked below the clothes”.

            • barsoap@lemm.ee
              link
              fedilink
              arrow-up
              4
              ·
              5 months ago

              Anime models are kind of notorious for drawing impossible clothes if you don’t guard against it in prompts and keep an eye out for them.

              To the point where I noticed that navel and impossible crotch folds before I noticed that extra arm.

              Realistic ones don’t tend to have that issue must be the training data, worst of all “impossible clothes” is not tagged reliably. Are those models able to draw proper clavicles? Absolutely. But they’ll also glue the collar to the skin because that’s how artists like to draw stuff like latex.