Don't worry, seeing the scene in pixiv, twitter/X, imageboards, Reddit, Instagram, RUclips slop lore/clickbait videos and so on, it will get even more worse with each year or so
@@CorvuSphere I think things that are accidentally horrifying are scarier than things that are purposefully made scary. So I think letting them do this accidentally might give some better results
As someone who's been working my *** of for the last 10 years to become an illustrator it's very depressing but this is the first ai video I've seen that aren't just ****** on and taunting artists
Oh good "question"! Note that there's a lot of debate about this between ML engineers, it's hard to confidently lean one way without more research. I like the idea of Nightshade more so than Glaze, but the effectiveness of both seems like it would come down to two things. First, they require widespread usage, ease of use and accessibility to be effective. Though given the archiving nature of the internet, already published work may be a lost cause, which is great for developing pruning methods to pick out bad samples. Second, as Nightshade/Glaze and generative AI tech improve, Nighshade/Glaze can become unusable if the only way to properly deter means there are noticeable visual disturbances to the human eye. In other words, it might not be future proof. I'm afraid this will eventually be the case, but I think widespread use would help the current state.
When you give the job of design over to uncreative technicians, uncreative but technically competent output is what you get. The main casualty of AI arriving will be the art itself. You have no idea how insanely soulless and mediocre mainstream art is about to become.
I've never used Dream Booth but looking at its paper, it looks like it's a decent method if you're in a situation where you don't have a lot of data. More specifically, it helps improve sentence context. However it does seem to be very sensitive to hyperparameters (the modifiable parameters whose values control the learning process) so you might have to play around with those. Automatic1111 is considered to be one of the best web UIs for stable diffusion models so that's a good choice. Let me know how it turns out if you try it!
Don't worry, seeing the scene in pixiv, twitter/X, imageboards, Reddit, Instagram, RUclips slop lore/clickbait videos and so on, it will get even more worse with each year or so
AI could technically be used to make body horror, because some AIs tend to be good at it (accidentally)
Good point, I wonder what happens when I actively try to make it horrifying...
@@CorvuSphere I think things that are accidentally horrifying are scarier than things that are purposefully made scary.
So I think letting them do this accidentally might give some better results
RUclips recomended this video ramdomly.
High quality, hope this gets around.
As someone who's been working my *** of for the last 10 years to become an illustrator it's very depressing but this is the first ai video I've seen that aren't just ****** on and taunting artists
I would appreciate your analysis on Nightshade and Glaze that is currently being championed by artists to protect illustrations.
Oh good "question"! Note that there's a lot of debate about this between ML engineers, it's hard to confidently lean one way without more research. I like the idea of Nightshade more so than Glaze, but the effectiveness of both seems like it would come down to two things.
First, they require widespread usage, ease of use and accessibility to be effective. Though given the archiving nature of the internet, already published work may be a lost cause, which is great for developing pruning methods to pick out bad samples.
Second, as Nightshade/Glaze and generative AI tech improve, Nighshade/Glaze can become unusable if the only way to properly deter means there are noticeable visual disturbances to the human eye. In other words, it might not be future proof. I'm afraid this will eventually be the case, but I think widespread use would help the current state.
@@CorvuSphere Thank you for your insight! I'm surprised that Nightshade is actually better than Glaze. Interesting.
When you give the job of design over to uncreative technicians, uncreative but technically competent output is what you get. The main casualty of AI arriving will be the art itself. You have no idea how insanely soulless and mediocre mainstream art is about to become.
Do you know if Dream Booth in Automatic1111 is any good for training AI?
I've never used Dream Booth but looking at its paper, it looks like it's a decent method if you're in a situation where you don't have a lot of data. More specifically, it helps improve sentence context. However it does seem to be very sensitive to hyperparameters (the modifiable parameters whose values control the learning process) so you might have to play around with those.
Automatic1111 is considered to be one of the best web UIs for stable diffusion models so that's a good choice.
Let me know how it turns out if you try it!
come on man, your videos are will, make better thumbnails, so people start to watch more, focus on that and the titles, videos will be more recommend