📣 Clarification on AI generated work

[joseph]

This site is for humans to share artwork ideated and made by humans using words ideated and written by humans.

AI can help as a tool, but majority AI generation is a no-go.

Notes:

• This is not a new policy, this has always been the case, however it needs stated officially.
• This is not up for debate and will not be changed. The staff agrees on this position.
• Sharing AI generated work won’t result in account restrictions, generally speaking, the thread will simply be closed and unlisted. That is to say, you’re not going to get in trouble, the content will just be removed

For example, using an AI translator to translate your thoughts is fine. They are your thoughts. Using AI to write a reply is not- you did not ideate those words. They are not human thoughts.

Sharing a work that used an AI retopology tool, where the concept and execution were made by humans, is fine. Sharing a work based on an AI concept that an AI made into a mesh is not. It’s not a human idea and it’s not human work on display.

Typing prompts into a generator is not human ideation, at least on BlenderArtists. Other sites may disagree- that’s fine, you are welcome to share your AI-generated work on those sites. The definition of human ideation for BlenderArtists is an idea originated and developed in a human brain.

[rogper (Rogério Perdiz)]

I couldn’t agree more! Well done!

[BlenderTimer (Daniel Roberts)]

1000% agree!

[MartinZ (Martynas Žiemys)]

I like the rule. It’s nice to only see what is done by actual Blender artists or at least humans who are attempting to become artists using Blender. With that said, AI is sort of slowly becoming a tool just like any other tool. It might become hard not to use it in the future, because who wants to be left behind? Personally, I do imagine it being used as a tool. There are even examples of that where AI is actually used to improve works. Now that particular example could probably be posted here and no one would mind, because a reasonable person would probably agree that it would follow the rules, but the line is sometimes not that clear and the decision is opinion based. Maybe it’s worth considering having a separate section on the forums for works that are done with the help of AI? I mean not a section for thousands of AI generated images but for people to post their work when AI is used to help make it where this rule could still apply but maybe a bit less strictly?.. Maybe not now, maybe if and when the hype slows down a bit in the future…

It’s still interesting to see how(and if) Blender artists may use AI to not only make meaningless AI slop.

[joseph]

@MartinZ

With that said, AI is sort of slowly becoming a tool just like any other tool. It might become hard not to use it in the future, because who wants to be left behind? Personally, I do imagine it being used as a tool. There are even examples of that where AI is actually used to improve works. Now that particular example could probably be posted here and no one would mind, because a reasonable person would probably agree that it would follow the rules, but the line is sometimes not that clear and the decision is opinion based.

It is quite clear:

@joseph

AI can help as a tool, but majority AI generation is a no-go.

Did an AI do 51% or more of the work? If so, it’s not allowed here

[rawalanche]

@joseph

Did an AI do 51% or more of the work? If so, it’s not allowed here

How do you quantify it though? Do you go by time spent? Probably not, otherwise nearly everything AI generated would pass given how fast AI generates things. But then you again need someone to subjectively judge how much effort/work individual parts of the workflow pipeline are.

It also contradicts your own original post, which states it’s not about the proportional amount of work but rather whether the work started with an AI generated idea/reference or not.

[joseph]

@rawalanche

It also contradicts your own original post, which states it’s not about the proportional amount of work but rather whether the work started with an AI generated idea/reference or not.

I never said anything like that. Don’t put words in my mouth.

The policy is this: site is for humans to share artwork ideated and made by humans using words ideated and written by humans. AI can help as a tool, but majority AI generation is a no-go.

You can nitpick at your leisure, but ultimately it boils down to- do you know what a majority is? Assuming you do, there’s no ground to nitpick.

@rawalanche

How do you quantify it though? Do you go by time spent? Probably not, otherwise nearly everything AI generated would pass given how fast AI generates things.

I don’t quantify it, it doesn’t need quantified. The policy could be not be more simple. If the text or image you are presenting is 51% AI generated (ideated and/or created) or more, it’s not allowed. Time doesn’t matter nor is it relevant. Nor is effort or pipelines or anything else.

[rawalanche]

@joseph

You can nitpick at your leisure, but ultimately it boils down to- do you know what a majority is? Assuming you do, there’s no ground to nitpick.

No, I genuinely don’t.

@joseph

I don’t quantify it, it doesn’t need quantified. The policy could be not be more simple. If the text or image you are presenting is 51% AI generated (ideated and/or created) or more, it’s not allowed. Time doesn’t matter nor is it relevant. Nor is effort or pipelines or anything else.

So what does this mean? Like if over 50% of the actual image surface, in pixels, is covered by something that processed by AI in any way?

I really genuinely don’t know that when I produce some imagery that AI has assisted me with, how can I evaluate whether I am below or above those 50%?

@joseph

I never said anything like that. Don’t put words in my mouth.

@joseph

The definition of human ideation for BlenderArtists is an idea originated and developed in a human brain.

Ok, so if you did not mean this as in reference or the source image/concept, than what did you mean? Because technically the prompts for the AI image generators originate, and are developed in a human brain.

[joseph]

@rawalanche

Ok, so if you did not mean this as in reference or the source image/concept, than what did you mean? Because technically the prompts for the AI image generators originate, and are developed in a human brain.

You know exactly what I meant, because I already said:

Typing prompts into a generator is not human ideation, at least on BlenderArtists. Other sites may disagree- that’s fine, you are welcome to share your AI-generated work on those sites. The definition of human ideation for BlenderArtists is an idea originated and developed in a human brain.

[thorn (BA Staff)]

@MartinZ

Maybe it’s worth considering having a separate section on the forums for works that are done with the help of AI?

People can post such things on other websites. Social media is filling up with that garbage; we don’t need it here.

@rawalanche

Because technically the prompts for the AI image generators originate, and are developed in a human brain.

There is nothing artistically skilled about typing “girl in a forest looking at a river”.

It’s hardly more involved than using that phrase in a google image search, and we don’t encourage people to download pics off google, run it through a filter, and post it as their own creation.

[thorn (BA Staff)]

@rawalanche

I really genuinely don’t know that when I produce some imagery that AI has assisted me with, how can I evaluate whether I am below or above those 50%?

Don’t use AI, and you’ll know the percentage.

[rawalanche]

Yes, I understand that part, that you consider output of text prompt based AI image generators to adhere to the rules. I completely get that.

But it’s not that simple. It’s not a matter of just taking that image and posting it directly here on BA. Obviously that’s bad. It’s about more nuanced use cases. Like using just parts of the AI generated images.
For example:

• Photobashing pieces of AI generated images together with real world photos, 3D renders/viewport screenshots, and manual brush strokes
• Using Img2Img instead of Text2Img, where you drawn the sketch yourself
• Using AI generated images as textures in 3D scene
• Using AI enhancers like magnific AI on your own 3D renders at the end

For more practical example. These are two renders I’ve made myself all from scratch and rendered back in mental ray days, 2009-2011, and their magnific-AI enhanced counterparts:

Orig:

Enhanced:

Orig:

Enhanced:

If I think about 50% rule, then by the image surface, yes, this won’t fly because the entire image is essentially AI. If I think about it in terms of if the image originated in my head, then yes, it will fly. If I think about it in terms of how would it would take me to manually reach the same detail and quality level as the AI, then again it won’t fly because AI did more than 50% of work.

You may probably say that this is a no brainer, that you can clearly see that I did most of the work, based on your feelings, but what about this example, where I fed my own 3D artwork into Img2Img model:

I did the exact same of work initially, doing everything from the scratch way before generative AI was a thing (that’s why I intentionally used ~15 year old examples of my work), and then in both scenarios, I used some sort of AI filter, just of different “strength”, but how do I know if whether that strength exceeded this arbitrary 50% mark?

@thorn

Don’t use AI, and you’ll know the percentage.

I am sorry, but comments like these add nothing to the discussion. I’d agree with you if the rule was simply no AI-involved art allowed at all, but I am just trying to clarify what the rules are.

[thorn (BA Staff)]

@rawalanche

I am sorry, but comments like these add nothing to the discussion.

Neither does this game of “but what is AI?” you’re trying to draw people into…

I fed my own 3D artwork into Img2Img model:

And you think that you created that final image? You did not.

If you don’t want to create art yourself - that’s fine, don’t. Use every chatbot you can find, you’re quite free to do that.

But don’t hang out on a forum for chefs, post pics of your takeout, and then claim that “I ordered this food, therefore I’m a chef too.” You clearly HAVE artistic skills, and it’s frankly sad that you’re deciding to walk away from it and let a chatbot do it instead.

[joseph]

@rawalanche

You may probably say that this is a no brainer, that you can clearly see that I did most of the work, based on your feelings, but what about this example, where I fed my own 3D artwork into Img2Img model:

100% of the output is AI-generated. 50% of the creation was human, 50% was AI. That gives us a 75% AI ratio. That’s clearly a majority

[rawalanche]

@thorn

And you think that you created that final image? You did not.

Did I claim that anywhere? No? The question is, did I create any of the slightly improved magnificAI ones? If your answer is “Yes”, then where’s the line between the two? If your answer is also “You did not”, then where do you draw a line between magnific AI and Optix denoiser?

@joseph

100% of the output is AI-generated. 50% of the creation was human, 50% was AI. That gives us a 75% AI ratio. That’s clearly a majority

Then I’d just update the rules to ban any AI-involved art whatsoever, because the formula you just came up with is nothing that can be derived from any logical framework.

[joseph]

@rawalanche

Optix denoiser?

Optix denoising is not generative AI. There’s no need for false equivalency.

@rawalanche

Then I’d just update the rules to ban any AI-involved art whatsoever, because the formula you just came up with is nothing that can be derived from any logical framework.

You’re certainly welcome to that opinion. It has no bearing on the policy. I’ve offered plenty of clarification, there’s nothing further to clarify

[thetony20]

@thorn

Don’t use AI, and you’ll know the percentage.

True, but as time goes on, especially from a production point of view, that is likely to be much harder to do, as in not use any AI.
I half suspect that in many ‘commercial’ operations, it may already be hard to do and if it isn’t now, then it likely soon will be.

That aside, how about this situation.
Now I’m no big fan of AI, but at this stage, for my Emilee character, I have used image generation. I wanted a ‘unique’ pattern for the skirt. So based on my already human established design ideas and look, I used AI to generate for me 10 image generation prompts for a seamless texture.
I then tried a number of those out with random seed values till I got a repeatable pattern that I liked.
I then used that as an alpha mask for a pattern on the overall skirt material.

That’s it, the total use of AI on the character at this stage. Now I think we would all agree the final render/animation, etc isn’t AI generated and unless I told you that the pattern was, no one would have even known that was AI made.

But, what happens if we take it to the next level.
Rather then just an alpha pattern, what if every material/PBR shader is AI generated? In effect every rendered pixel is ‘AI’, however the character concept and even the modelling was all still human made by hand.

We still good?

Or the character design is human, it’s then sculpted, but then the final model is totally AI generated/retopology (yeah, bit less likely at this stage, given it’s unlikely to get all that perfect. Mind you, so many people don’t do animation, so for a still image, a perfect mesh isn’t really needed).
And again, all PBR materials are AI created. Now, not only is every pixel AI made, but so is the 3D model it’s all sitting on.

Personally, it’s not something I ever plan on doing, but at the same time, I’m not working to a deadline in a studio or have a boss yelling for a poster or something for a client.
But I can see it coming, as much as I may not like it, the line is going to get more blurry, as AI becomes the new kitbash, asset model library, material library, etc.

[rawalanche]

@joseph

Optix denoising is not generative AI. There’s no need for false equivalency.

Perfect, so you drew the line between magnificAI enhancer and Optix denoiser.

What would you then say about AI upscalers? Those work very similarly to enhancers, in that they “make up detail” based on something they’ve been trained on, but their purpose is not creative enhancement, but just technical image quality enhancement - similar to Optix. Would posting magnific AI upscaled (not enhanced) work get it removed?

@joseph

You’re certainly welcome to that opinion. It has no bearing on the policy. I’ve offered plenty of clarification, there’s nothing further to clarify

Well it’s just impractical. Because the rules essentially imply that what constitutes too much AI use is at the discretion of the moderators, and will be judged on per submission basis. That there’s not much one can do in advance to figure out the outcome of their submission if their use of AI is borderline.

[thorn (BA Staff)]

This strategy game of “but what is X” is same one that went on for hours/days in the “no politics rule”, and during congressional hearings on the PMRC labels, and of course “what is p-rn.”

There is always an exception to such rules, but one will note that the original artwork that started this entire conversation wasn’t a complete human-generated artwork run through a denoiser. It was an AI concept, and I believe an AI mesh.

So there’s no need to get into the slim margin of “what about", nor create example images with the complete goal of making the example a blurry grey line.