[ 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.44845878 [View]
File: 140 KB, 1280x720, maxresdefault (21).jpg [View same] [iqdb] [saucenao] [google]
44845878

>>44845227
>>44845322
That isn't quite true. Even picture AIs trained on photos struggle with hands. The actual reasoning is because hands are extremely complex features with endless permutations. A face is a face, and a face looks pretty similar from most angles. Faces are also featured prominently in most artwork or pictures, creating a easily scrapable data set. In contrast, hands are delicate. Hands tilted just 10 degrees looks like a dramatically different object, and in artwork hands are splayed in all sorts of different angles - there's no consistency there. AIs are taught that all of these different images, the open fingers, the closed hands, the fist, the palm, these are all one object - the "hand" - which it tries to parse by jumbling all of these concepts together.
>>44845322
>>44845189
Which Elira LORA is that?

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