• SasquatchBanana@lemmy.world
          link
          fedilink
          arrow-up
          3
          arrow-down
          2
          ·
          1 year ago

          I am a bit impressed they were able to integrate the message in. I wouldn’t have thought an AI to be able to do this yet

          • 🔍🦘🛎@lemmy.world
            link
            fedilink
            English
            arrow-up
            7
            ·
            edit-2
            1 year ago

            It’s super impressive to me, and getting this kind of fidelity is honestly brand new. You can include an image in an AI image generator to serve as an outline. In this case a blank image with bold text. It’ll fill in the prompt using the reference as best it can (in this case, a bunch of kittens/puppies).

            Prior to a few weeks ago, I’d only ever seen this using a posed figure–basically a stick figure–to generate the prompt’s subject around.

          • general_kitten
            link
            fedilink
            arrow-up
            5
            ·
            edit-2
            1 year ago

            I think the way stable diffusion works makes this kind of message imprinting quite easy to implement.

            the wikipedia article has some good pictures

            I would imagine it works by inserting the text on some of the first steps so the ai has the text as it’s seed instead of just random noise.

      • ChaoticNeutralCzech@feddit.de
        link
        fedilink
        arrow-up
        4
        ·
        edit-2
        1 year ago

        I think the prompt is not much other than “puppies” and “kittens”. Major, middle and minor features of the image can be controlled individually in some AIs (they can be differentiated using a Fourier transform or Gauss convolutions and fed into different discriminators) so I think:

        • major features (scenery) are controlled by the prompt (grass or couch)
        • middle features (text) are a source image that the AI is punished for straying from
        • minor features (details) are controlled by the prompt (faces and fur)

        Or it’s just Stable Diffusion that starts with a text rather than random noise.

      • jballs@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        3
        ·
        1 year ago

        Not sure what this prompt was, since I didn’t make this one. I linked the site I used above, and they’re pretty simple to do but need a few tries to get a good one.