[ 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.6445608 [View]
File: 638 KB, 1372x551, AIPrompterTestKangaroo.png [View same] [iqdb] [saucenao] [google]
6445608

>>6443011
>image shitters is a far greater threat to the value of artists and their reasons for sharing their work
AI shitters are image remixers, not real artists. The AI does all the work & uses img2img, using a few images that were created by the real artists, then adds slight changes here & there. You could classify all AI art as "edits" because they are modifications of original art pieces.

Look at pic-related as an example. The AI clearly used images of Kangaroo Jack then modified a few things here & there. It is an edit.

>> No.6439224 [View]
File: 638 KB, 1372x551, asdfasdd.png [View same] [iqdb] [saucenao] [google]
6439224

>>6439198
I both draw and play with AI. Gives me an awareness of how limited AI is and how to use it to maximum capacity. Control comes in 2 forms, limiting yourself to text alone is using the AI at the lowest level.

Most people are unaware that the portion of the AI interpreting your text only has 0.8Billion parameters. That's teeny-tiny. It's not going to understand most things, barely understands english to begin with. See image, how it improves as you add more parameters. It's possible a larger model could begin to separate concepts to succeed the cubes/ball test.

>AI art program still could work without the need of drawing everything
Going to disagree, it does need imagery to unlock lot of untapped potential unable to be accessed through text alone. Doesn't need to be good, needs only a rough indication to guide the AI, but these rough indications need to showcase the fundaments of art and understanding of rendering. This is the best chance to make up for the AI's lack of understanding of perspective (AI needs to train on those Vandruff vids) and value work.

I know people talk about 3d mesh mapping with AI and blah blah blah, but this overhead is going not only slow down the process by adding extra steps, also going to hit a ceiling where AI users just don't have the capacity to load the models to begin with.

>>6439217
Where AI could get to the point AI people promise everything will be 'fixed' is going to get incredibly expensive. Going to need some pricey GPUs, $/hr to rent or a rack of them at home. I've laughed as people talk about how they run their GPU's all night just to get the result they want, where I've spent 5 minutes blobbing color on a canvas to get the results I'm aiming for in minutes.

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