[ 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.6897030 [View]
File: 528 KB, 470x657, 1693170070691609.webm [View same] [iqdb] [saucenao] [google]
6897030

>>6897021
who is this plagiarizing?

i'm sure you think that there is some pieces of artwork out there that this has "stolen" from.
...otherwise, how could this be possible? machines can't learn after all.
and this is why you're a retard.
:)

>> No.6869770 [View]
File: 528 KB, 470x657, 1694439296147878.webm [View same] [iqdb] [saucenao] [google]
6869770

>>6869631
>So, basically, it has noise saved
no. neither.
no noise is ever saved. this also shows a complete lack of understanding for how this works.
what these models actually do is REMOVE noise. so basically it tries to see the prompt you provided among the noise.
you can almost say it's a "cheeky trick" where it tries to hallucinate things from within the noise, the noise just represents a nigh infinite amount of starting possibilities.
it gradually removes the noise until nothing but an image is left.

-so during training the model tries to recover the exact training images from noise. (as it gets things wrong, it tries to correct itself. this process adjusts the model. this is the learning part)
-but during inference (when it makes images) it tries to recover a image from within the noise, except that image doesn't actually exist and is only represented by your prompt. so the resulting image is just everything it has learned about the tags during training.

>and keywords attached to that noise
i'm not super knowledgeable about this particular part as it's a bit hard to understand for a novice. but as far as i understand, keywords are tied to the weights in the model. basically when it sees the tag cat, the nodes connect in a certain way, and that is the part that is being trained. the node. i.e. the "neurons".
so here again, it's no individual image or even multiples of images being saved. the only thing remaining is a giant network of nodes that behaves a specific way when met with the input of "cat".

>then it reconstructs an image using various tagged noise based on the keywords used giving you an output that tries to match the keywords you used?
>How is that any different from photobashing or collage then?
like i explained above. because it is not using any specific image, but only its own learned representation of the tags.

look it up yourself how these things work if you don't believe me.

>> No.6859521 [View]
File: 528 KB, 470x657, 1674293742212911.webm [View same] [iqdb] [saucenao] [google]
6859521

Is AI learning and applying or is it stealing and copying?
If it is stealing, how do you think it works?
discuss.

i'd say it's very much art related. you can't just pretend it's /g/ because you you don't want to hear it.
this board is called artwork SLASH critique.

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