Currently our rules do not specify for or against AI postings, simply if something “makes you go awww!” then it is welcome. Given that more than a few users have expressed that AI pictures should be limited to AI communities, do we as a community want to restrict AI posts?

  • j4k3@lemmy.world
    link
    fedilink
    English
    arrow-up
    0
    arrow-down
    1
    ·
    1 year ago

    I understand and struggle with the idea of possible dishonesty. TBH I only saw this post in my local feed and commented about why it doesn’t make sense to ban this. The negative culture against this new tech is more dangerous than even the tech itself. This is why I posted and have kept watching this thread. The images generated are based on real images. It is challenging for a beginner to learn how to prompt for both the composition and photorealism. Composition is mostly done with positive prompting; saying what you want to see with the right collection of words and phrases. Photorealism requires both positive prompts about lighting, focus, and image characteristics, but also a negative prompt that removes large portions of the dataset, like CGI, cartoons, anime, etc.

    I don’t know of a single reason I would ever waste time on SD trying to generate images to post here. The reason I am trying to say it matters is because everyone needs to be aware of what is and is not possible in this space and stay up to date on it. The tech is moving extremely fast. The there are a lot of very subtle issues that creep into Stable Diffusion outputs. The more detailed and specific the prompt gets, the harder it is to prevent these small errors. After I played with the technology for a few days, I quickly learned to spot many of the errors. Learning this skill right now, to be able to spot the subtle AI errors in images may be really important in the near future. After learning this, I can spot a lot more images that are text2img generated and I would not have known otherwise.

    This technology is not a fad. It is not going away. It is a very powerful tool. Calling it “AI” is rather dumb. It is not AGI (general) in any remote sense. This is probably the most disruptive tech since the internet of the mid 90’s or capacitive touch tech enabling effective smart phone interfaces, or crypto currency making billionaires.

    It doesn’t matter where you get familiar with text2img nuances, you need to see the almost good, good, virtually indistinguishable, and truly indistinguishable potential to know what is really possible here. You really need to see the level of complexity that is currently possible in a prompt. There is a limit here. If you know these types if limits, you will have a much better sense of truth in this space. Everything visual should be met with skepticism now. Relegation of this tech to the engineers gives them extreme power over you now and even more in the near future. Please take the time to learn about this and be familiar with it.