Open weights AI generated text-to-3D and image-to-3D
also supports hybrid inputs (give it an image and text)

Try free online, no account needed:
https://huggingface.co/spaces/tencent/Hunyuan3D-2

download models for free:
https://huggingface.co/tencent/Hunyuan3D-2/tree/main
*also includes a separate model called "Paint" for creating and texturing UV maps from a mesh input
*The output mesh is a trimesh object, which you could save to glb/obj (or other file format).

LICENSE:
https://huggingface.co/tencent/Hunyuan3D-2/blob/main/LICENSE

>>1005982
tl;dr?

I was planning to pick up character sculpting to start selling 3d printable figurines. Is there no point anymore?

i'll spin this up once there's a comfyui implementation. the cherry picked results look okay, but textures on some examples look like overcooked shit (like what happens if you crank cfg scale up to 10 on diffusion models)

>>1005984
I'm not a lawyer but
>2 - Rights: Free to use, modify, and redistribute, subject to adherence to the Acceptable use policy
>3 - Distribution: give credit, provide copy of license, notify third party of any changes
>4 - Commercial: if licensee has more than 1 million monthly active users of all products and services, additional terms apply (need express written approval from Tencent)
>5 - Rules: follow all applicable laws
>6 - IP: don't pretend to be Tencent. users are responsible for the content and derivatives they make
>7 - Warranty / liability: no support or updates. no warranty.
>8 - Termination: The license is active until terminated.
>9 - Jurisdiction: disputes are exclusive jurisdiction of Hong Kong
Acceptable Use Policy:
>use prohibited in European Union and South Korea
>don't violate any law or regulation
>don't harm yourself or others
>don't exploit or harm minors
>don't defame or harass others
>don't generate content for the purpose of harming electronic systems (viruses / hacking)
>don't impersonate anyone without consent (deepfakes)
>don't violate social ethics
>no terrorism or extremism
>no military use
>no malpractice or fraud

>>1005986
This discussion has been had in other areas like programming and art. Here's my take:
1) You have to know the AI and how to use it. AI can only get better from here. Like with all other revolutionary technologies in the past, everyone either adopted the tech (some sooner, some later) or failed to compete and lost.
2) Focus on complementary skills to the AI: Find out where the current AI fails and focus on those skillsets. If AI does something exceptionally well, it should be a low priority.
Think of AI as a tool - the most awesome goddamned tool of this century. The time is coming soon when we either have to adapt to it or lose to it.

Wow, very cool!

>>1005986


Don't listen to this fucking faggot >>1006032 . AI is not a tool, it's a replacement. There's no "using" fucking AI because it's a replacement advertised specifically so. It's marketed to non creators who want content to consume and corporations who want to save money. It's not for the creators.

"Learn AI". What a fucking joke lmao, as if prompting will be a real job or even relevant enough. The point is to cut costs but cutting labor. That means massive amount of job loss, that means a hyper competitive industry shrinks even more. There's no room for the creator to "adapt to AI", it automates the process. Either it isn't as good as these people are saying and it's useless, or it will continue to get better rapidly to the point you aren't needed in the pipeline. 3D thankfully is safer from this because there isn't enough data like there is with 2D images to train it. But I don't imagine in 20-30 years anything that can be done on a computer will still be done by humans.

>>1006086
They are in the denial phase.
>It's just a tool and requires an artist
Those faggots are coping hard right now.

>>1005981 (OP)

Is there any merit in using an AI mesh as a way to quickly establish proportions in 3D? You'd sculpt/model the final thing yourself but you can use the AI mesh as a way to essentially skip the blockout phase.

only errors.

What does those thing do?

>>1006272
AI Gacha isn't going to beat Shift+D or Procedural Generation. Learn to make use of what's already out there.

>>1006272
Blocking your model out is barely any work though. People already use basemeshes for sculpting, and this is just creating worse basemeshes, you still need to sculpt over the things it gives you. So far not that good.

It's about the same as trellis and neither is good enough yet.
>It only turns one input image, into one 3d model. When I'm sculpting, I want to turn five images into a new model, not an exact replica of one image.
>The AI needs the input image to be a realistic full body view on a transparent background. It doesn't understand cartoons.
>It depends on AI generated input images.
This guy shows it more https://www.youtube.com/watch?v=g-ThznwPxJ8

>>1006107
>>1006032
Basically what >>1006086 said, with some caveats. "Learning AI" is the biggest cope there is, we had AI for various fields and uses for 3+ years now and outside of programmers nobody has embraced it. Image AIs are only used by no-draws and the small artefacts or just the way the colors and contrast work give it away eventhough it was worked on for so long and has so much time to improve, and like predicted once people get over the novelty, they came to see all AI content indiscriminately as slop. Even if it looks objectively good regardless of the tiny artefacts, the moment your brain notices that artefact and registers it as AI picture it will become slop in most people's minds, and same will happen with these models. So look at what 2D AI has done and expect similar to take shape in 3D, maybe with some more caveats because it still cant do good rig or anything.

Overall what I am trying to say is that this will most likely be used by cheap motherfuckers making slop, or used to make background stock models like 2D was used for making textures and stock videos. It will cause a lot of people to lose jobs, and make the market more competitive, but it wont kill it outright, just like it didnt kill drawing and how NSFW artists still make pretty much the same money on Patreon as they did before AI, despite there being infinite cheap AI porn.

>>1006344
I'm positive towards AI because I don't work in 3D so my livelyhood does not depend on it. But what's the point of models like that? You can do that in 10 minutes using the snakehook brush and you still need to completely sculpt over it.

>>1006357
Now you can hire a prompt engineer for half the price.

>>1005981 (OP)
Hmmm looks like slop

>>1006351
This anon has great points. I believe people will reject slop more than embracing for the sake of being too easy/similar to other stuff.

lmao it's so over for 3d artists, programmers and everyone else. we are all done for. you are literally training yourself for no future. some AI will replace you within a few years and you'll be back at mcdonalds putting fries in the bag.

it's over boys, we had a good run. gonna order some rope.

>>1006379
They said same thing for literally every single breakthrough tech but somehow only more jobs emerged in the long run.

Not too bad but too many tris

>>1006379
Programmers maybe but definitely not for (all) artists, if you produce slop with no personal touch that people don't care about you will obviously get replaced, which would mostly be industry jobs

>>1006385
The amount of tris doesn't matter much these days, considering how good Maya's auto-retopo is these days. It's literally a couple clicks, and you have a serviceable mesh. Not perfect of course, but definitely usable for many use cases.

>>1006387
I'm noob and never tried Maya, should I? Does Blender have anything close to auto-retopo?

>>1006388
I'm not experienced enough with Blender to comment on that side, but you can try the auto-retopo feature in Maya's month long free trial, to see if it compares to whatever option blender has.

>>1006384
>>1006386
nah it's totally over for artists. just as with 2D it will start with AI assistants who greatly simplify tasks, this alone will make the job so much easier that any jeet can do it for 5 bucks an hour. then AI will start to do the actual work. you want a house? just prompt it, AI will shit one out. want a different style? prompt it to change up styles. don't like the color? prooompt. need better detailing in a certain area? just select the area, prooompt away. oh what's that, you want a certain section of the wall to be be damaged? select it, prooompt for damage, cycle through the results. UV unwrapping, baking or any other meme is handled automatically by the AI.

this isn't even a meme or exaggeration. all of this will happen, and it will happen really fucking fast. like up to 5 years at most. we are completely and utterly done for.

>>1006390
I agree overall, but I doubt animators and directors will be replaced that fast, maybe animators in 10 years. Modeller and even rigger positions will be among the first, and 5 years doesn't sound too unrealistic. Director will be the last job to remain in the industry, and the competition will be fierce, with everyone remaining becoming one, and having their own product. That is until it too gets overtaken eventually.

>>1006344
It can run on a single GPU, that's huge. Also these are just first steps, imagine what they will do in 5 years.

>>1006390
>>1006392
Normies will start hanging aibros in the streets before that happens

>>1006388
>Does Blender have anything close to auto-retopo?
Yes, but the default Blender quad remesher is not good. Most people will use the Quad Remesher addon instead, which is really just zremesher if you're familiar with zbrush, same guy.
https://exoside.com

>>1006031
>>don't violate social ethics
COOMERS STILL WINNING

>>1006344
>>1006357
yeah i guess in 10 years we will see a lot of good shit

>>1006393
>>1006499
>imagine what they will do in 5 years.
>10 years we will see a lot of good shit
progress isn't inevitable
technology doesn't necessarily scale predictably

>>1006518
They said we're going back to the moon.
I guess our government has done good prompts up and ready.

>>1006518
Current AI is barely at the Sputnik 1. So yeah we have like at least 5 years. Except AGI and androids with human rights are way closer than useful space travelling.

>>1006357
>But what's the point of models like that? You can do that in 10 minutes using the snakehook brush and you still need to completely sculpt over it.
NTA, I'm using AI to make t-pose models for posing units in blender for my miniature games. They look like shit compared to an actual sculpter's output, but I'm not an actual sculpter, I'm a rando, and at 28mm I can hardly tell the difference anyway. It lowered the bar for getting into modeling from "learn to be an artist from scratch, here's a blender cube, fuck you" to "here's a perfectly serviceable model to pose how you want and sculpt just the bits you don't."

>adjust any of the settings
>aborted task
great program

>>1005981 (OP)
>>1006344
I've noticed that a lot of these AI 3D Model generators always produce squished humans and I think that has to do with the fact that the volume of the input data is nowhere near the size of a human. This is why most of these models advertise with chibi characters because their proportions more closely fill the rectangular volume compared to a tall humanoid character.

>>1005981 (OP)
>error
>error
>error
>shitty result
>error
>error
>aborted task

amazing

>>1005981 (OP)
it can make porn!

Ehh...

>>1007588

>>1007589

>>1007590
Last one.

If there were multiple image support, there could be more potential than a singular image.

>>1006392
>maybe animators in 10 years. Modeller and even rigger positions will be among the first
It's literally the opposite imo. Modelers will be the last to go as the need for high quality clean meshes will not go down. Animation and texturing will be the first to go, because it's the easier to replicate

>>1006272
It's a subpar blockout, and if this seems useful to you, you're not making anything worthwhile from the mesh you're getting. But once you get to that point that it's better than nothing - and you'll get to that point and easier quicker without it - it's not useful. So, why use it?

You can see it here pretty clearly >>1006344, all of the shapes are wrong, even if you can appreciate that an attempt was made, so why not make the attempt yourself a few dozen times, and quickly learn to be better than it?

This is a blizzard model. The texture quality and strange geometry in game makes me think that this is a touched up AI model. Can anyone confirm?

>>1010201
Hard to tell but definitely possible. Textures and mesh look pretty much indiscernible from actual human work albeit poor so if it was touched up it's pretty much impossible to tell. The baked in lighting is rather ai looking but I'm not super familiar with overwatch assets. Bigger giveway to me is just the base idea and concept of it, including pose and proportions, looks fairly uncanny, not something a human brain would design, made out of inconsistent patterns

>>1010226
> the base idea and concept of it, including pose and proportions, looks fairly uncanny, not something a human brain would design, made out of inconsistent patterns
So it looks procgen?
Because AI only generates stuff similar to what it was trained on, that is, human work.
One of the big weaknesses of AI is exactly that it isn't able to break conventions and proportions.

>>1006390
>and it will happen really fucking fast

You people said the same back in 2022. It is already the year we should totally be done for, but here I am, still creating.

>>1006606
You should have said thank you for those 12 free triangles in the starter cube along with the miscellaneous light source

>>1005981 (OP)
Industry insider here, just dropping by to let you know that all artists will be replaced in economic viability within the next 2 years

We can theoretically do it now, and studios who do simpler work already have, but we still need a couple technical artists and a senior artist to coordinate them as well as to orchestrate the gens for 3d (for now).

If you're still learning traditional 3d, you're doing it with the expectation that there will be no money or possibly even no audience in it even if you get really good.

>>1012042
you need to be 13 to post here, anon

>>1006032
You are completely wrong, ai is designed to replace you, not to complement you.
It's retarded that you think that you will need to learn to use ai, in the future ai would be implemented in everything, you will not be necessary as a middle man. At this point ai makes the bulk of the work but it's slightly complicated for the majority. As soon as it becomes easier to use you are done. Everyone, even a child can give a prompt, There is no meaningful skill in that, even if you don't know how to explain yourself you can ask ai to make a well written prompt from your abstract idea.
There's no place in the future for artists using ai, not using ai is the future.

>>1012042
Real industry insider here, ai will kill the industry more than already is.
We as an industry create slop for mentally ill people, but now it's going to be ai slop for mentally ill people.
Instead of an LGBTQ writer with daddy issues we have an ai trained with millions of LGBTQ writers with all the mental illness you can find.
Mix that with tasteless non artists using random generated art without direction and we have the future of our Industry.

>>1012304
> There is no meaningful skill in that, even if you don't know how to explain yourself you can ask ai to make a well written prompt from your abstract idea.
Your abstract idea: gimme a fun game looking good
AI produces Fortnite
You play Fortnite. Happy.
You will never be able to produce something like this:
https://www.youtube.com/watch?v=ScglDSi9KUs

One of things about AI is that most often you don't really have the information you need to even prompt what you would arrive at if you performed some of the work manually.
You discover concepts and ideas during the process of producing, and that is a very big part of result.
AI sure will be everywhere but even in the most endgame scenario human work is always required simply to even figure out what you want and how you want it.
Otherwise you can't produce novel prompts, novel training sets, etc.

>>1012313
People who use ai have already given up on art, you'll find they typically believe that it's impossible to create something new or original. It's just an outer coating for whatever grift they have

With code, programmers use it to write small code sections for mundane stuff instead of spitting out the entire program. You can tell it the logic and what you expect, then read its output and fix the stupid stuff it does. It's only beneficial to people who already know how to code, because newbs use it do all their uni assignments and never learn to code. They can't recognise bad code or how to debug their errors, so their skill cap is the AI's skill cap. When you have to write anything more complex than a uni assignment to make a sudoku solver, or work on a large existing codebase, you struggle because AI can't do everything.

I'm new to /3d/, so take this with a grain of salt, but instead of inputting a 2d picture and generating a 3d model, I think it'll help with smaller steps in the process. I'm using AI to get ideas for stylised environmental reference art. It's really good at this. While you can't get the proportions for character references, you can get good ideas. Obviously, the underlying skills for reference art are needed, but it'll help the process. It would significantly reduce the market for artists as it's doing for programmers, as all the low-hanging tasks can be done faster with it. The problem is that this is the worst gen ai is ever going to be. It's going to keep improving at a fast pace.

Man all of this AI shit is completely useless to me since I'm only interested in highly stylized and 2D looking 3D which all of these generators are absolutely terrible at, they can only make funko pop toy aesthetic looking shit.

>>1012311
>Your abstract idea: gimme a fun game looking good
That's not even close anon, I'm not fan of ai but I can tell you that most humans, even children can form more strong and precise sentences than "make pretty game".
Ai is made to replace you, most gamers will replace you, they consume the most retarded slop possible because they are used to eat shit.

>>1012320
>It's going to keep improving at a fast pace.
And soon most jobs will be consumed by ai

>>1012332
Not really. It will be more precise than that, but it will still be not precise enough to specify every new unique style that is physically possible,
because many of currently unexplored unique styles can only be produced as a byproduct of messing around with things manually.
This is also true about things other than styles, with minor variations on what manual intervention is key to discovery of thing in question.

>>1012332
>>1012333
>>1012350
Yeah, I'm not saying people will not eat shit, that's ofc true and will probably become worse.
What I'm saying is that there are still good objective reasons to do some things manually even in advanced AI era.

>>1012311
Yeah, it cant make something like this, but since the average gamer wont have the idea to even play this, they wont miss it. You are not missing on the potential hypothetical world of advancements if digital never became real because you dont even know it could even be real. And if you think it could be real you will be able to prompt AI do it for you. World where Hyperdemon does not exist isnt one where gamers wish Hyperdemon existed, its a world where the person who would make it does not get money from it, thats it.

>>1012294
>>1012306
the entire concept art team where I work got laid off months ago, this shit is happening and unless you're specialized in a niche that's hard to replace it's genuinely over for most of us in the coming years

>>1012362
> World where Hyperdemon does not exist isnt one where gamers wish Hyperdemon existed, its a world where the person who would make it does not get money from it, thats it.
How on earth do you come to conclusions like this? It's not an industry-funded game, no one was sitting on salary doing it. It's an indie game made by a single person.
Not a single gamer knew they wanted it before its author made it. And yet it was commercially successful and got decent recognition, especially for a single-person indie game.
If every industry game will be AI driven and levels of "generic" increase there, games like this will only become more popular than they are now.

>>1005986
>i was thinking of how i can sell things
>is there no point

boohoo AI took away your get rich quick idea. you should be making art because you enjoy it, not to make a buck. if AI ever does get to the point that it replaces a lot of jobs, i'm sure the economy will equalize things and a lot of goods and services will be a lot cheaper.

>>1012833
>if AI ever does get to the point that it replaces a lot of jobs, i'm sure the economy will equalize things and a lot of goods and services will be a lot cheaper

This is not what's going to happen. It's going to remove 30-50% of the workforce which won't have jobs. If things get cheaper it won't matter because people won't have jobs. If UBI is implemented it will only be enough to keep everyone barely scraping by while tech giants live like kings. The reality is AI is there to destroy the human workforce.

>boohoo AI took away your get rich quick idea. you should be making art because you enjoy it, not to make a buck.
Nigga he's trying to just make a living you faggot. People aren't even going to be able to see human made art with the flood of AI slop.

>>1012374
What's a niche that's hard to replace?

>>1012875
Flower sculptor

>>1012875
Anything that can't be summarized in a prompt

www.neural4d.com/research/direct3d-s2
https://huggingface.co/wushuang98/Direct3D-S2
It's worst than Hunyuan 2.5 for sure, but seems to do better with details. Probably has holes and etc. though.

>>1012902
Does it run on 10GB of VRAM?

Kind of annoying how the more these model generators "advance" the more they rely on designs with small stout proportions, and they all have this cheap toy aesthetic. Not like it matters to producers, as long as it's fast they're put this shit in any pipeline they can get their hands on.

>>1012910
They use point clouds and sign distance fields to create the surface. So it will always have this rounded look until they can crank the resolution even higher without using too much hardware resources.
I still like using these models as a starting point, kind of like how photobashing is used in digital painting.

>>1006263
beautiful

>>1006272
Blockout is pretty fasr once you've learned it, you're probably just gonna take as much time fixing teh AI mesh. Plus if you want to skip blockout you could just get a ready made base mesh