Willenator Unscripted

Because Normal Is Overrated

Tag: AI

  • More AI Slop

    I’ve been going through my old posts and fixing the posts that were imported over from Tumblr recently, and I came across a post from 2022 where I tried to have the Dall-E image generator create me an image of Penn and Teller performing a magic trick with Donald Duck. The result was a little disturbing:

    An image showing two guys on stage with a photo-realistic looking cartoon duck on the table. The faces of the two guys are distorted. Their hands are shaped weirdly. The duck that is supposed to resemble Donald Duck has a top hat. There are starts on the wall in the background that look weird.

    I got to wondering how well the image generators would do now with a similar prompt, so I hopped onto Copilot and ChatGPT and tried to get them to create the image. No luck. Donald Duck is trademarked, and both didn’t want to create something with famous people in them.

    Copilot even suggested that I create something with magicians inspired by Penn & Teller and a cartoon duck, so I tried that. It still rejected the prompt because the magicians it was thinking of creating would resemble Penn & Teller. That amused me, so I said “how about you just create an image of two magicians on a Vegas stage performing a trick with a duck.” It still rejected it because it would be too close to Penn & Teller and responded with “you can tell me to make fully original magicians.” So I did.

    It looks like the system blocked that attempt because the request was still interpreted as involving real, identifiable people, even though we shifted to original characters.

    I can’t generate this image under those constraints.

    This is probably due to the fact that the conversation I was in with Copilot had me talking about Penn & Teller, so it was stuck on Penn & Teller even though I tried to move away from it. I’ve seen other people talk about this challenge that if you tell an AI to create an image of animals but no elephants, the AI will always add an elephant.

    I did try to create a the image using a new chat thread, and this was the result:

    An image of two magicians (one male and one female) performing a magic trick with a cartoon duck. The two people are on a stage with bright lights above. There are a few audience members seen at the bottom. There is a magicians hat floating in the middle with the duck coming out of it and the guy is holding another magicians hat as well. The lady is holding a wand.

    Meanwhile, I decided to hop on X and see what Grok would do since that AI doesn’t have many guide rails like the others do, and this was the result:

    Two men in suits stand indoors beside a person dressed in a full Donald Duck costume. One man smiles while holding a conductor’s baton, and the other stands just behind the character, also smiling. Donald Duck wears his classic blue sailor outfit, black bow tie, and blue hat. The scene appears to be at an event or gathering, with all three posing for the photo.

    I don’t know who the magicians are, but that is definitely Donald Duck.

    The funny thing about this post was I wanted to comment on how much image generation has improved over the last four years, and I ended up going down a different AI rabbit hole of fighting with the AI because of the restrictions that are now in place.

    This amused me and helped me waste 30 minutes of my day.

  • Western Digital Article Comment

    Thanks a lot, AI: Hard drives are already sold out for the entire year, says Western Digital

    Sigh. Gues this means we will see prices go up on hard drives now.