Just kidding, mostly. I’m working on a presentation for my company about AI, and one of the things we want to do is illustrate the risk of deep fakes, and to do that our idea is to generate an AI image of one of the managers at a Taylor Swift concert or something like that totally out of character for him. It’s playful, not meant to be malicious, and I’ve got buy in from upper management. I’m also not looking to do an actual deepfake, it would be enough if the image had a strong resemblance. The problem I’m running into is ChatGPT and Bard, the two I thought to try, will not generate a description of a person (in this case the managers headshot), and I’m not great at describing people, so I’m kind of at a dead end.

Any advice appreciated.

Also I recognize that while I swear my intended use is completely innocent, the answer here could be used in unethical ways, so I completely understand if mods want to take this down.

  • AtHeartEngineer@lemmy.world
    link
    fedilink
    English
    arrow-up
    5
    ·
    10 months ago

    If you have a decent computer, and are tech savvy, stable diffusion with a face swap extension is probably your best bet

  • Dagamant@lemmy.world
    link
    fedilink
    arrow-up
    4
    ·
    10 months ago

    There are lots of “deep fake” apps available for phones that can make short videos of a person or swap their face onto a celebrity in a short scene. You could use stable diffusion to train a model on photos of your boss and have it generate images of him in different situations. You could use you could use Sad Talker to animate a photo into a video of him saying something. You could use tortoise-tts to generate audio of him saying different things after training it with a few minutes of his voice.

    A lot of what I mentioned is free software but requires a lot of learning time, effort, and a good video card to get working. I think what you need to do is clearly define what you want to demonstrate and how to make it have an impact without being malicious.

    Some examples I would use would be a faked audio file of him asking you to work late or congratulating you on your presentation, images of him having coffee with former presidents or celebrities (preferably dead ones), or images of him fighting dinosaurs or aliens. Keep it light hearted but let it show what tech can do.

    Most of the things I mentioned are also plugins for automatic1111 which is a free program for using stable diffusion and other AI tools but, you can probably find apps for your phone that will require less effort.