📣 Clarification on AI generated work

[rawalanche]

@marcatore

but probably I’m ignoring something

What you are ignoring is that when a human looks at other human’s work, it’s called inspiration, but when a machine does it, it’s called stealing.

You’ve identified the problem: AI with an empty database would not be able to produce anything, just like human with empty brain. Human can look around and start drawing something, just like AI can look around and start drawing something, it just doesn’t have eyes in the same biological sense as humans.

Here’s a link to a thought experiment I posted in the previous, now closed thread, with a more detailed explanation:

Rules about posting content created with the help of AI on BlenderArtists

Ok, but then consider following case: You have a person that grows up on a lone island, never seeing much more than sea, sand, stones, sky and palm trees in their life. Maybe a fish here and there. Once they mature, you visit them, put a pencil and paper in their hand, and ask them to draw a “scifi spaceship concept art”. They will have obviously no clue what you are talking about. Most of fictional art these days is derived from fictional art of other people, through the inspiration. If we co…

[Luigi.exe]

I think you’re massively over-blowing my statement in a comically slippery-slope. Seems like the people in this thread love the slippery-slope fallacy and it’s getting kind of old. We are discussing generative AI, not asset packs, not templates. Stop changing the subject.

Geometry nodes still requires a form of creativity, even if it’s not standard creativity. It still requires legitimate human input, and again, does not require a dataset consisting of the entire illegally-scraped internet to function. People’s misunderstanding of what is meant by “generative AI” is starting to get a bit frustrating.

Other sites like Unvale have completely banned generative AI, and that site is doing just fine.

[Luigi.exe]

Okay, no. Inspiration is not the same as a model being fed an exact image that is permanently stored on a database that is never warped or distorted.

I’m so tired of people comparing generative AI to a human brain. They are not the same. Generative AI is not capable of critical thought. The type of model we are talking about is the kind of model that will adamantly assert that “strawberry” has 2 r’s in it. Generative AI cannot understand the data it is being fed, it cannot critically analyze its data. Tell it enough times that 2 + 2 = fish and it will regurgitate that uncritically.

Models like ChatGPT are basically advanced text predictors, machines of averages, machines built to mimic human expression. That is not how the human brain works.

[sozap]

@Luigi.exe

Need references? Ever heard of photo-bashing? Internet search? Real life? Other media? Character creators? Libraries? Maybe connecting with other people?

A baseline to work with? Asset packs. Templates.

Models to kit-bash a scene? Asset packs, volunteer work, or god forbid, commissioning an artist.

Well, maybe we don’t envision those AI tools the same way. Personally I didn’t look into it yet, but I’m looking forward at some point to try stuff like iterating over a design/drawing I did to look for ideas/improvements.

Same goes with lighting or rendering, it could be cool to let the AI have a go at your piece and pick a few ideas here and there.

But maybe we should phrase the rules a bit differently. Maybe what is a bit controversial would be to use AI to improve a final render to a point where there would be a huge difference between the original and the AI improved version ?

While I tend to be against over-abusing AI and too many shortcuts especially in personal work I think those tools are part of an artist toolbag. At some point in history photoshop, 3D, kitbash, photobash and all the tricks artist can use were considered cheating against some kind of old school technique.

I find also particularly weird when stating :

@Luigi.exe

Creativity is not something that should be automated.

Because we use 3D which automates all the perspective and light calculations, the vast majority of 3D artists would be completely lost if they where asked to trade their 3D software with some pencil and paint to do the same renders.

And I’m pretty sure when 3D came out some people saw it as pure evil that should be banned.

Anyway, I’m not that much in favor of AI but I’m kind of curious about how talented people will use it, and I’d be very happy if we can find a fair rule that allows it’s use while preventing people from over-abusing it…

[Luigi.exe]

Because we use 3D which automates all the perspective and light calculations, the vast majority of 3D artists would be completely lost if they where asked to trade their 3D software with some pencil and paint to do the same renders.

That is not what I meant and you know it.

I have already explained that generative AI takes over the creative process, not enhance it.

With 3D lighting, guess what? You still use your brain and figure out if the lighting looks good or not. You still choose where the lights go, their intensity, their color, how many of them there are. That is not the same as generative AI just making a setup for you with no critical thinking on your part. I genuinely do not understand what is so difficult to understand about that. Generative AI creates finished outputs, not help someone make theirs.

[rawalanche]

You’re right, an AI is not a human brain. But your post highlights an interesting paradox.

In one breath, you’re criticizing AI for having a superhuman ability (a photographic memory), and in the next, you’re criticizing it for having a sub-human ability (it can’t think critically). So the AI is flawed because it does some things better than humans and some things worse?

This is why the whole debate about the AI’s ‘nature’ misses the point when it comes to copyright. The focus should be on user responsibility.

Long before generative AI was a thing, you could do the exact same thing. You could hire an artist and tell them to copy a specific frame from a movie. The artist doesn’t have perfect memory (unlike the AI, which you criticized), but that doesn’t stop them. They just open an external database - Google Images - and copy it from there.

So the artist used a tool to access a database to replicate work at my direction. If I then go and use that image commercially, who should be liable? It should be me, the person who commissioned and misused the work. Not the artist, and definitely not Google.

It’s the same thing with AI. If a user gets an AI to reproduce copyrighted stuff and then uses it improperly, the blame falls on them. The tech is new, but the idea of user responsibility has been around forever.

[Helmut_S]

@Luigi.exe

With 3D lighting, guess what? You still use your brain and figure out if the lighting looks good or not. You still choose where the lights go, their intensity, their color, how many of them there are.

Playing devil’s advocate here: you can do much of the same with genAI. While in 3D it happens sort of bottom up, it’s more like clever directing of what’s already present, in genAI.
There are plenty of tricks far exceeding simple prompting, to manipulate lighting for instance. Just have a look at ComfyUI and the huge amount of available tools and workflows. I’d say there’s the same level of (technical) creativity involved that’s demanded by typical 3d work.

That doesn’t change the fact however, how all genAI is created on a foundation of theft, and how genAI is used as a corporate honeypot that does not play in favor of the user.

[Luigi.exe]

Okay, you still don’t seem to understand it.

Does a human require a massive database of literally stolen content, said database that massively pollutes the environment to exist in any meaningful capacity? No?

And yeah, no duh generative AI has strengths and weaknesses. Really don’t get what you’re trying to counteract. Generative AI was not built to think critically or be self aware: it was made to mimic human expression via statistics and word association. Obviously it’s greatly flawed.

And you’re little scenario with the artist: again, they’re only copying one single image. Generative AI models like Stable Diffusion require millions of images to function even remotely. And even with an artist copying it, it’s still got human flaws and error and would never be a perfect recreation. That artist’s brain does not have an exact copy of the image they can just spit out at the behest of a prompt in a matter of seconds. That’s not how human brains work.

Stop trying to compare a massive warehouse of stolen media to a human brain. It’s not going to work. Not just that, but nobody brought up copyright. What is with you people and changing the subject?

Helmut_S
Playing devil’s advocate here: you can do much of the same with genAI. While in 3D it happens sort of bottom up, it’s more like clever directing of what’s already present, in genAI.
There are plenty of tricks far exceeding simple prompting, to manipulate lighting for instance. Just have a look at ComfyUI and the huge amount of available tools and workflows. I’d say there’s the same level of (technical) creativity involved that’s demanded by typical 3d work.

Also, thank you for proving my point that generative AI provides nothing novel. Exactly as I said earlier, you can do exactly that with bases and templates made by people without generating assets from stolen data. You can also do this in things like video games and creators.

[sozap]

Well, I think everyone agrees here that a completely generated image has nothing to do here, since it’s made with an AI and not blender.

Now give me what you think of these use cases :
A / I’m looking for a background / sky for my 3D render, I use AI to generate a couple of image and I pick the best one, I then edit it further in a Paint app before using it as a texture into my scene.

B/ I’m working on a concept / drawing of a character , I use an AI to get variations over that design, and pick a few ideas in different variation that I incorporate in my final design before turning it into 3D.

C/ I’ve modeled a cool scene, but wondering a bit where to go with the lighting, I use an AI to take a viewport render and render different mood and light scenarios, which I use as a reference alongside movie frames to set my final lighting.

D/ I want to work on a scene but what I want is a bit unclear, I use AI to generate a couple of pictures to get an initial moodboard for inspiration, that I further refine with real world reference to build a final concept.

E/ I want to model a complex object, I use AI to generate a rough 3D model that could serve as a blocking / 3D reference , temp mesh to test out if everything works before taking the time to model the final mesh.

My point is, I don’t see any issue with using AI if it’s done cleverly, I think everyone that is going to over-abuse it will get a result that will scream AI and I agree with you that’s something we should avoid. However, I’m sure a lot of talented artists will use it as a tool in a way that you probably won’t notice it if they don’t tell you , is that something you’re still having an issue with ?

[rawalanche]

You’re bringing up several points here, so let’s tackle them one by one.

First, you accuse me of changing the subject to copyright, but then you immediately change the subject to the environmental impact of data centers. The concept of a “database of literally stolen content” that you brought up is a copyright argument. It’s central to this whole topic.

Let’s go back to the artist scenario, because I think it proves my point.

You say a human is different because they don’t need a massive database. I disagree. If you ask a human artist to perfectly replicate a movie frame they saw once, they will fail unless they have a photographic memory. To succeed, they absolutely need an external database like Google Images, which contains billions of copyrighted images. So yes, a human does require a massive database to perform a high-fidelity task without failure. They just access it differently. If you call the AI’s learning “stealing”, then to be consistent, isn’t an artist studying copyrighted images on Google to learn a style also “stealing content”?

This brings me to what I think is your real issue. You pointed out that the human copy would have “flaws and error,” wouldn’t be a “perfect recreation,” and couldn’t be done in “a matter of seconds.”

This tells me your problem isn’t with the ethics of the process itself (since a human can do a similar thing). Your problem is with the AI’s efficiency. You don’t seem to have an issue with the principle of an artist copying something, but you have a huge issue with a tool that can do it faster, more accurately, and at a larger scale. The objection isn’t about morals, it’s about the speed and productivity of the machine.

Lastly, you criticize the AI for not having critical thought. You’re right, it doesn’t. But why is that a flaw in an image generation tool? Its job isn’t to be a philosopher, its job is to create images based on prompts. If the results are good enough to be useful, then it’s succeeding at the task it was designed for, regardless of whether it “thinks” like a human.

[Helmut_S]

@Luigi.exe

Stop trying to compare a massive warehouse of stolen media to a human brain.

It’s not, though. It’s a neural net trained on a massive warehouse of stolen media. This point has been made again, and again, and again.

It’s not going to work. Not just that, but nobody brought up copyright.

Good point worth mentioning once again … however no one seems to care.
I wonder whether that’s being considered a thing of the past already, or … because the copyright-agnostic “download stuff from the net for my own works” culture is so widespread in the art community already … and here we’re circling back to assets, base meshes and so on … I’ve seen so many characters built upon clearly unlicensed base meshes coming from DAZ, iClone etc … also this stupid “great artists steal” meme, nothing but a cheap excuse for the lazy, “efficient” industry copy cat (anyone ever heard of rumors how the original version’s first part was “BAD artists COPY …”?).
I keep wondering whether the ground for treating intellectual property the way the AI tech community treats it, has been paved by parts of the artist community too.

Also, thank you for proving my point that generative AI provides nothing novel.

That’s not what I did.

[Meshmonkey]

No weapons on featured row is a rule? That’s new to me and apparantly to Bart as well

Renaissance Crossbow

Hi Blenders! Inspired by historical Renaissance crossbows with ornaments I challenged myself to recreate one of them with a personal touch. I wanted create an hunting scene, the details are made by hand and I tried to create a balanced composition to avoid visual clutter, I made them with the help of a lot Renaissance illustrations. Hope you like it! [Cover_noLogo] [detail_1_B] [detail_2] [Orto_1] [silhouette] If you wanna see more about this: https://www.arts…

[BG_Division]

Yeah, when I still was a big cheese eater user (aka a Regular ) I stopped voting that kind of post for featured row even if the image was insanely good. No context, no additional images or nothing about the workflow used (even just a few details), it was a no go for me.

[Kurtis]

@Ace_Dragon

Though we need to be careful, or we otherwise face a reality of the main Blender community declaring the use of some features within Blender itself as off limits.

I’m sorry, but this is dishonest slippery slope argument and completely misses the mark of the conversation. No-one here is suggesting banning tools within Blender itself, either now nor in the future; the discussion is surrounding the use of generative AI within people’s works, and just the use of generative AI within people’s works.

The entire conversation on here since about asset libraries, and tools within Blender is moot. That is a giant red-herring; there is a stark difference between using AI generative prompts to create works, than using pre-made by human assets.

@Ace_Dragon

It can also be said that enforcing the idea that creativity can’t be automated can even be a gateway for a prohibition on the use of Geometry Nodes, procedural texturing, simulation tools, and even path tracing. Why not have everyone then go back to modeling, shading, and lighting, everything by hand and painting every texel, nice and slow and destructive, just how it was done in the early 1990’s in the name of ‘craft’? We might all want to have a version of Blender pre 2.x on hand just to make sure we are safe. The very idea of extending the new policy beyond obvious AI generation to implicating existing technologies is a major can of worms that this community will come to regret opening.

This is all a different kind of automation. In every single tool you’ve mentioned there, there is a human element - a person controls what procedural textures to use, how to use normals with them, how to set the colors, the maps, the scaling, etc. Even in simulations, you get so much control. If you didn’t, what would be the point? AI takes all of that away - it brings the creativity of the “artist” down to a single ruddy prompt - and even then that prompt is a response to ‘trained’ AI works (as has been established.) This, specifically, is where the objection is.

[Luigi.exe]

Now give me what you think of these use cases :

I’m not sure why you’re so keen on debating me when I’m not the one with the say here.

Now give me what you think of these use cases :
A / I’m looking for a background / sky for my 3D render, I use AI to generate a couple of image and I pick the best one, I then edit it further in a Paint app before using it as a texture into my scene.

That’s akin to what Shadversity does. No, wouldn’t allow it. Still used generative-AI to make assets, even if you did edit it slightly. The source still does not change.

B/ I’m working on a concept / drawing of a character , I use an AI to get variations over that design, and pick a few ideas in different variation that I incorporate in my final design before turning it into 3D.

Still no. Still used generative AI get the base concept.

C/ I’ve modeled a cool scene, but wondering a bit where to go with the lighting, I use an AI to take a viewport render and render different mood and light scenarios, which I use as a reference alongside movie frames to set my final lighting.

Still no, generative AI was used.

D/ I want to work on a scene but what I want is a bit unclear, I use AI to generate a couple of pictures to get an initial moodboard for inspiration, that I further refine with real world reference to build a final concept.

Nope.

E/ I want to model a complex object, I use AI to generate a rough 3D model that could serve as a blocking / 3D reference , temp mesh to test out if everything works before taking the time to model the final mesh.

And nope.

I will admit that sadly, there is a decent chance an artist may wind up using generative AI as a source of inspiration on accident, though. But deliberately generating via these models? That’s using generative AI in your work. Really not as gray as you think it is with what you listed. I doubt Joseph or Bart would even be able to tell. Just don’t use it deliberately.

Like I said, everything you listed can be done without generative AI. You can get inspiration elsewhere, you can use filters in things like Photoshop to test lighting, you can photobash concepts, etc.

rawalanche
First, you accuse me of changing the subject to copyright, but then you immediately change the subject to the environmental impact of data centers. The concept of a “database of literally stolen content” that you brought up is a copyright argument. It’s central to this whole topic.

This tells me your problem isn’t with the ethics of the process itself (since a human can do a similar thing). Your problem is with the AI’s efficiency. You don’t seem to have an issue with the principle of an artist copying something, but you have a huge issue with a tool that can do it faster, more accurately, and at a larger scale. The objection isn’t about morals, it’s about the speed and productivity of the machine.

Yeah, no, you could not be more wrong, oh wow. Uh yeah, no, I hate the plagiarism machine with a burning passion.

Okay, no, I’m not doing this. I’m getting the vibe you’re not arguing in good-faith. I’m tired of running in semantic circles with you. This is the exact kind of rule-lawyering I was talking about that the complete ban would avoid. People like you going over a 500 page law book on the philosophical nature of a literal pathologically lying plagiarism machine made by billionaires to steal jobs from the working class to save money while also wrecking the environment and polluting the internet en masse.

I am more okay with an artist tracing art versus generative AI because the human gave enough of a care to actually trace it (and also doesn’t consume a town’s-worth of energy to do so!). NO, before any of you crop up saying, “So you’re fine with tracing???” no. Stop. No, it’s not okay, but if I had to pick and choose between generative AI and some shmuck copying an art piece, I’m going with the shmuck. At least when someone traces, they improve their skills and learn something. At least a beginner can use tracing as a learning tool as long as they don’t claim credit for what they’ve traced. Generative AI does not do that. Your comparisons continue to and will continue to fall flat.

And honestly, I’m getting the feeling that several people here do not have faith in the people who’ve run this site for over a decade to make good judgement calls over what is and isn’t generative AI. This site has existed just fine for over a decade, Bart has not steered us wrong for that long, and I doubt this will be different. Bring on the total generative AI ban, I welcome it.

[thetony20]

@Kurtis

The entire conversation on here since about asset libraries, and tools within Blender is moot. That is a giant red-herring; there is a stark difference between using AI generative prompts to create works, than using pre-made by human assets.

But is it?
So if ‘generative AI’ is used to make the concept and then ‘generative AI’ is used to create the models based on that concept and the ‘artist’ then drops it into Blender, adds some lights, a bit of fog and hits render.
Is that OK… based on OP post, it isn’t.

However, if the same person does a google image search, finds a photo they like, downloads some matching or near matching model assets, drops it in Blender, adds some lights and a bit of fog and hits render.
Are you saying that’s all perfectly fine?

Other then the tools used, whats the difference?

@Helmut_S

It’s a neural net trained on a massive warehouse of stolen media.

So seen this a few times now. What if it isn’t stolen media. What if say Photoshop/Firefly is used, which I believe (tho I could be wrong), that the AI used for that, Adobe trained using their stock photo library.
So since the data isn’t stolen, is the result now all OK to use, or does the actual source make no difference.

If the source makes no difference, then see my points above, which in turn would make both of those cases basically the same and hence they are either both OK or not OK.

[joseph]

This is all wildly off topic. This thread is about policy, not asset packs or AI copyright or whatever else. Since no one has asked real questions about the policy in 50+ posts, I take that to mean the policy is adequately explained and clarified. To that end, I’m closing this for replies