正在加载视频...

视频加载失败

Oh wow, I wonder how it did that.

72,452 次观看 • 1 个月前 •via X (Twitter)

30 条评论

Reid Southen 的头像
Reid Southen1 个月前

Guys, how did it do this!?

Yaroslav Golubev 的头像
Yaroslav Golubev1 个月前

Like this:

Michael 的头像
Michael1 个月前

Surely they didn’t scrape tons of 3d models for the ai to generate the mesh. Surely

Jscott 的头像
Jscott1 个月前

Pretty sure it just downloaded this exact asset. The seems look like shitty photogrammetry.

Leon 的头像
Leon1 个月前

Epic's acquisition of Sketchfab and Artstation and making every course free for a year without an ulterior motive seemed too good to be true back then. It really was too good to be true.

K. S. A. G. 的头像
K. S. A. G.1 个月前

The sloppers are discovering online repositories of free 3D models inside the training vector space. Hilarity ensures

GutenLeinad 的头像
GutenLeinad1 个月前

You are making a fool of yourself. I have tried it myself and yes it can do it on its own from scratch while using references (like any modeler would). There are even people showing the whole process. I can only guess this is reverse AI-psychosis.

Reid Southen 的头像
Reid Southen1 个月前

Do it with something that has zero 3D models of it out there already, let's see how well it performs.

GutenLeinad 的头像
GutenLeinad1 个月前

People are literally using AI generated images as reference that are turned into models and those certainly didn't exist before. I also did my own tests with things that certainly don't exist anywhere else and it did it. As long as it is hardsurface it can produce decnt results.

Reid Southen 的头像
Reid Southen1 个月前

Okay, then show me, I asked to see how well it performs when there aren't 3D models of something already, you claim you've seen it perform well, show me.

GutenLeinad 的头像
GutenLeinad1 个月前

Here you can see it live creating a bat. Here a random room. There are a million out there. It can do hardsurface pretty well and the topology is decent and it struggles with organic shapes.

Reid Southen 的头像
Reid Southen1 个月前

Lmfao, come on man, be serious. And that room is represented by a million generic sci-fi environment models on CGTrader, Sketchfab, Turbosquid and elsewhere.

GutenLeinad 的头像
GutenLeinad1 个月前

Now you are switching arguments after proving to be false? I was fair enough to not just show the best results because obviously there are plenty of scenarios where it doesn't do well. This discussion was about you claiming it would just grab other models.

Reid Southen 的头像
Reid Southen1 个月前

You didn't prove anything to be false, there are plenty of bat models out there, and millions of generic slick sci-fi interiors, lol. Find something actually unique that it can model effectively.

anon11472 的头像
anon114721 个月前

It’s hilarious to me that the AI models figured out, “Hey, you know I can just download this right? I don’t need to make it again,” long before the AI users did AI models: I can get whole scenes for free off sketchfab and just toss it in blender Ai users: fukkin magic bro

Inspired Taste 的头像
Inspired Taste1 个月前

👇👇👇

Ady 的头像
Ady1 个月前

How tech bros feel after spending $5 worth of tokens and 30 minutes making Astra generate a 3D asset that was already free

Mariko Sanogo 的头像
Mariko Sanogo1 个月前

Wow, I can't believe it there are more versions of a popular sculpture. Man, it's pathetic; the only consolation left is thinking they might be copying it, but that's not even the case.

Felipe_Miura✨ 的头像
Felipe_Miura✨1 个月前

I'm not crazy, right? It's #him

Eliottx 的头像
Eliottx1 个月前

Well yes because it is well known that a human who has raised in a black box he knows how to imagine the triumphal arc in 3d. we are so special and superior

Void 的头像
Void1 个月前

Hmmm i wonder

Arthur Bouffard 的头像
Arthur Bouffard1 个月前

given the output 3D topology im not sure these existing models play a direct role in the result. the ai agent writes code that then generates lines and curves from a precise vertex placement. from how it works it seems to understand dimensions from images and research, and can apply those into 3D space. it does gather textures from the internet and adds them to materials if you ask it to. if the topology matches exactly then yes it probably did use the model in its training set

Dave the Horse 🇨🇴 的头像
Dave the Horse 🇨🇴1 个月前

They are saying How did the AI made the reliefs? Bro you just put a BW image of the design on a displacement node and apply the modifier 😭😭😭😭 fucking tech illiterate

Brad 的头像
Brad1 个月前

Pretty terrible photo I snapped and prompt lol but worked for 32m and cranked out a render.

chs 的头像
chs1 个月前

I can't believe it made the same Arc de Triomphe as those others. What a copycat!

Nicolas Ribeiro 的头像
Nicolas Ribeiro1 个月前

Again the performance argument dont hold up. But worse, the second part don't hold up neither: let's imagine you modelise a new building and it will also something close with your new thing. You are so strangely delusionnal about AI. There is so much to speak rather than that.

Rishabh 的头像
Rishabh1 个月前

Wait this is stealing....

-… 🔫 的头像
-… 🔫1 个月前

doesn’t work like that, you’re a ignorant.

Will Sidebottom 的头像
Will Sidebottom1 个月前

Hahahaha.

Friendly Neighborhood Butlerian 的头像
Friendly Neighborhood Butlerian1 个月前

It’s a Scooby doo mystery.

相关视频

Oh geez oh dang I wonder how this obstacle will be resolved.
0:43

Sensitive content

Oh geez oh dang I wonder how this obstacle will be resolved.

SeraF

131,341 次观看 • 12 天前