[ 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.

/vt/ - Virtual Youtubers

Search:


View post   

>> No.68438733 [View]
File: 1.44 MB, 1280x1536, 1704250385015214.png [View same] [iqdb] [saucenao] [google]
68438733

>>68436067
eh, that's closer to a symptom than a problem
even now we still haven't totally abandoned eps-pred since a lot of resources are still on it, and people don't really know the difference between eps-pred and v-pred
furries are still figuring shit out about SD1.5 and the number of problems that would need full retrains (so unfixable) just keeps getting higher
as I mentioned, soonTM the anime model trained by furries will resume training, which means we'll finally have another base model and we will be able to ditch NAI cancer

>>68437340
>It's always the tards that don't clean up what they gen that seem to pass off clearly gen'd pics as handmade art
yeah, most of the AI that people see posted online is from troglodytes who gen at lowres and slap a patreon link on it when it's clearly base NAI with no care about the quality

>It won't ever surpass an actual human making art
I really don't understand people making this argument, especially when it's not in an emotional response because you felt attacked by it.
It's not doing anything by itself since it NEEDS conditioning provided by a human (text, image, or any controlnet tranlation layer), so it's just a tool limited by the person behind the screen
In the greater scheme that statement holds even worse since ML is still a blooming field, and even though a lot of it is just "stack more layers and compute", a lot of research is made into new architectures and improvements upon old ones. People don't realize how fast the field is moving even with shit like the subsequent DALL-E releases right in front of their eyes.

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