And that was based on my own render, too.
from artstation just a moment ago: "we have introduced a “NoAI” tag. When you tag your projects using the “NoAI” tag, the project will automatically be assigned an HTML “NoAI” meta tag. This will mark the project so that AI systems know you explicitly disallow the use of the project and its contained content by AI systems. "
1 Like
The NoAi tag is pointless, it’s up to developers to honor it and they don’t. Besides, the databases have already been made anyway. This is a false flag move by ArtStation
6 Likes
That’s exactly my stance about these AI projects.
I am all in favour to tools that can (to some extent) assist artists perform tedious tasks more easily (Cascadeur, CEB Easymocap Monocular, ChatGPT, Photoshop Neural Filters, LazyNezumiPro, …)
But I am not a fan of generating out-of-the-box final digital paintings thanks to Stable Diffusion webUIs or generating 3D models thanks to text-to-3D generators.
2 Likes
Wait, if they don’t have any ties or agreements with those AI companies, (like they also claimed) how do they plan to develop an integration with their training software? This is obviously a lie.
2 Likes
Yeah. They do that. 
hopefully the continued pressure will get an actual response
3 Likes
You already have AI help in that. It’s called Google Image Search. I know it’s sounds so mundane, but it is AI. We are used to it now, but think about how much it takes to get from “skull” typed into a web page to getting you skull images (many of which are not even tagged with a word skull, or that the word is not mentioned there). And that’s not taking into account similar images (when you click on one there you get bunch of similar to that one) or reverse image search.
4 Likes
You can’t have one without the other IMO.
“Corpo-speak”
Doubt it… and I actually don’t think they should bend to the pressure: their platform, their rules. They listened (hard not to), they made up their mind in some way. It’s not like there is no alternatives to hosting portfolio on artstation and that artists are forced to do so.
Btw. food for thought. From our point of view (artists) this whole development seems like a bad idea. But artists are tiny fraction of the society. What about other parts of human race? I think if we expand the view from just us, to the whole society there is way more positives for tools like midjourney than the pain caused by the need of adaptation to changing industry will cause.
2 Likes
I did not try cascadeur yet. But what I noticed looks interesting.
It bugs me that there could be growing artistic AI-Bypass. With artistic AI-Bypass I mean, comparing what human did is trivial compared to the results he gets.
The essential process to work/decisions/creation is mostly done by an AI.
Imo, artistic skills are something valuable. A way to handle AI-Bypass might be to include the process, how something is made. If AI used, to be honest where it got used.
What is the policy of blenderartist.org about AI scraping is there anyhting to prevent or is it a free for all like all the other site.
Thanks.
There’s nothing anyone can do on this website or any other to prevent images being scraped for AI databases- noAI meta tags are entirely voluntarily honored by developers, and they openly and consistently make it clear they do not care to honor them
4 Likes
Thanks for the clarification.
Now that i think about it the best way to avoid getting scraped is to pass your artwork through an ai that does nothing but it should be enough the be categorized as ai art and the bots will avoid the piece since they avoid feeding back their turd into the machine.
Remember who owns Art Station though, I do not know if Epic will really change course unless there was mass-scale uninstallations of Unreal 5 as everyone moves to Godot. A bonus, a lot of hard-drives and internet connections will breathe a sigh of relief.
2 Likes
Silicon Valley corps are deploying AI as parasitic agents that feed on society.
The pathology looks like this:
-Collect intelligence from the masses without our knowledge or consent. The collection continues even when we overtly raise our voice in opposition. No “opt-outs” allowed
-Inject that crowd-sourced intelligence into proprietary AI systems that subvert our interests, income and welfare
This is one big blood sucking monster, but because its fun and exciting and provides a modicum of benefit…we fail to see the larger threat.
Something tells me we the public at large won’t receive even a fraction of the profits, despite being responsible for 99.9% of the intellectual capital…the all-important library of knowledge through which AI learns.
1 Like
The question was whether there is a technique capable of achieving retopology.
What we are seeing with image generation and large language models are the results of roughly a decade of intensive research where a huge amount of resources were spent figuring out what works and what doesn’t. A lot of neural network architectures have been tried out, tons of training techniques have been refined. Tasks were found which work, datasets created for them, improvements being made on those, new tasks found, … . It was a ridiculous effort leading to what we are seeing right now. So much research is being published, about language and image models in machine learning that it is impossible to read all papers about it.
When it comes to handling mesh data with neural networks, there is pretty much nothing going on. Everyone is working on voxels and point clouds because those formats are a way better fit for neural networks.
It turns out that working with vertices which are connected to build faces is a very difficult format to handle for neural networks (for arbitrary amount of vertices and faces). I am confident this is going to be solved in one way or another, but there is no solution in sight right now that I am aware of.
3 Likes
All those rely on mesh data and as far as I know, there is pretty much no ongoing research for any of those.
This builds on techniques which have been refined for many years where papers have constantly been published, because people have been working on that problem.
As of today, pretty much for every promising AI solution, you find years of research work leading to it with plenty of published papers.
Edit: Just as a quick illustration. This is the work from one researcher (and collaborators over several years): https://www.youtube.com/channel/UC82J8WSryFmdTJuO7lzv79A
If someone had a look at the papers, it would be easy to find other researchers and results who have been working on that too, including plenty of working even making those results possible.
Nothing like that exists for mesh based solutions.
2 Likes
@DeepBlender
I see, I was too quick with my hope that AI might generate a base mesh from a sculpt.
Writing the note I remembered that there was an AI assisted way to warp a finished topology around a model.
Searching for it I forgot the ‘right words’, I get tons of false hits on Google.
Yes, there are warping kind of solutions which use a predefined topology, like this one:
There are others, if I remember correctly also for other tasks.
Unfortunately, generating a new topology seems really difficult to do.
One idea I had was to UV unwrap a sculpt and bake the position to each pixel. This would be some trick to pass mesh data to a neural network in a format that is known to work. I can imagine possibly having multiple of those, all differently unwrapped with varying edge seems might be feasible starting point. Even if that worked, it would be very far away from something that could be used with a practical value.
Any higher profile case of this to point to? I’ve seen the overall attitude for sure, but I haven’t come across to devs publically coming out to state this, even though that’s what they have been doing.
2 Likes