[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/ic/ - Artwork/Critique

Search:


View post   

>> No.6684031 [View]
File: 231 KB, 540x1727, 20230526_194056.jpg [View same] [iqdb] [saucenao] [google]
6684031

>>6683867
If it's only trained on your work then it's actually just useless. See picrelated. They can only output interpolations of pictures they've been fed, therefore needing millions and billions of images to have enough interpolations that either don't look like Lovecraftian horrors and are different enough from the base images to be able to pass as something new.
Essentially all it is is automated, gacharolled photobashing, with all the pitfalls of copying/plagiaristic workflows plus the inherent randomness of image generation.

>> No.6680979 [View]
File: 231 KB, 540x1727, 20230526_194056.jpg [View same] [iqdb] [saucenao] [google]
6680979

>>6680930
You are correct.
Essentially, all that a model can output are the images it was trained on + all possible interpolations between those images. It can shuffle and move things around too, I believe. But the end result is extremely similar to that of photobashing. The best example of this being the Sad Bench Keanu Reeves in armor generated by Midjourney. It's much harder to catch that happening with the anime stuff because there's such a huge amount of it, however I have seen a few examplss of artists' signatures appearing, gatlrbled but still recognizable as that of a real artist.
As others have called it, it's a data laundering machine.

>> No.6676767 [View]
File: 231 KB, 540x1727, 20230526_194056.jpg [View same] [iqdb] [saucenao] [google]
6676767

>>6669828
>There is an infinite amount. The amount of datasets is not finite, it is infinite.
*laughs in interpolation*
The jig is up, Sam. The latent space is static and its outputs are finite. There is no magic. There is no "mixing styles". There's only interpolated images all the way down.

>> No.6674916 [View]
File: 231 KB, 540x1727, 20230526_194056.jpg [View same] [iqdb] [saucenao] [google]
6674916

>>6674777
The thing is even if a human tries as hard as they can to ape someone else's style, there will always be something distinct abouttheir own art. The way they make strokes, the way they perceive, the way they think. People can't help but be unique. When lerning from others a human will always add something of their own.

Image generators do not. They can only interpolate between imges in their dataset. They add nothing, everything that they're capable of outputting is fixed from the outset- all it would take for someone to "steal" your "work" is to get the combination of seed & prompts that you used. There is absolitely nothing unique about you or your "art". You can photobash a bunch of other artists' work all you want, but you won't add anything or have your own style.

Really, that's the limit of what image generators are capable of- photobashing and shooping. No, that's not literally how they work, I already understand that just fine you massive fAIggots- but the end result is in effect the same. The image of Sad Bench Keanu Reeves in a suit of armor that's been floating around here on /ic/ is a great example of this.
You will never be unique. You will never have a style. You will never be an artist. You're just a cog in the machine, a cipher, easily replaceable, with no value or voice of your own.

Navigation
View posts[+24][+48][+96]