Top
Best
New

Posted by participant3 4/3/2025

An image of an archeologist adventurer who wears a hat and uses a bullwhip(theaiunderwriter.substack.com)
1503 points | 898 commentspage 3
armchairhacker 4/3/2025|
You're allowed to draw IP and share your drawings. You're allowed to screenshot and photoshop IP. You're allowed to sell tools that help others draw and photoshop IP*. You're not allowed to sell these drawings and photoshops.

I don't see why an AI can't generate IP, even if the AI is being sold. What's not allowed is selling the generated IP.

Style is even more permissive: you're allowed to sell something in any style. AFAIK the only things that can be restricted are methods to achieve the style (via patents), similar brands in similar service categories (via trademarks), and depictions of objects (via copyrights).

Note that something being "the original" gives it an intrinsic popularity advantage, and someone being "the original creator" gives their new works an intrinsic advantage. I believe in attribution, which means that if someone recreates or closely derives another's art or style, they should point to the original**. With attribution, IP is much less important, because a recreation or spin-off must be significantly better to out-compete the original in popularity, and even then, it's extra success usually spills onto the original, making it more popular than it would be without the recreation or spin-off anyways. Old books, movies, and video games like Harry Potter, Star Wars, and Sonic have many "recreations" which copy all but their IP, and fan recreations which copy even that; yet they're far more popular than all the recreations, and when Warner Bros, Disney, or SEGA release a new installment, the new installment is far more popular too, simply because it's an original.

* IANAL, maybe there are some technicalities, but in practice this is true.

** Or others can do it. As long as it shows up alongside their work, so people don't see the recreation or close derivation without knowing about the original.

lmm 4/3/2025||
> You're allowed to draw IP and share your drawings.

No you're not, not in general. The copyright holder has the exclusive right to prepare and distribute derivative works.

> You're allowed to screenshot and photoshop IP.

Again, no, not in general.

> You're allowed to sell tools that help others draw and photoshop IP*.

Sort of. You're allowed to sell tools that might be used for those purposes. You're not allowed to sell tools as "for that purpose" or advertise those use cases.

Smar 4/4/2025||
I guess country of residence has a lot of relevance here...
lmm 4/4/2025||
Not really, post-Berne convention.
tjpnz 4/4/2025|||
>I don't see why an AI can't generate IP, even if the AI is being sold. What's not allowed is selling the generated IP.

Where I live Studio Ghibli are known to be active on C2C marketplaces looking for counterfeit goods. If you were to list a home pressed Totoro pillowcase it would be taken down, sometimes proactively by the marketplace. From that perspective I struggle to see much discernable difference given OpenAI are offering a paid service which allows me to create similar items, and one could argue is promoting it too.

rescripting 4/3/2025|||
If I pay for ChatGPT and ask for copyrighted images, is that selling the generated IP?
alphan0n 4/4/2025|||
You are responsible for the output, just like any other tool.

If I use a copy machine to reproduce your copyrighted work, I am responsible for that infringement not Xerox.

If I coax your novel out of my phones keyboard suggestion engine letter by letter, and publish it, it’s still me infringing on your copyright.

If I make a copy of your clip art in Illustratator, is Adobe responsible? Etc.

VanTheBrand 4/4/2025|||
What if the ceo of xerox went on social media and promoted copy machines by showing how you could use them for infringement?
alphan0n 4/4/2025||
Is that what is happening in reality?

It seems that all of the big players in the industry are perfectly fine with disallowing output that infringes on copyright.

skydhash 4/4/2025|||
The analogies fail because the copyrighted material were not used for creating the copy machine, Illustration, or (maybe?) the keyboard suggestion engine. If LLMs were produced ethically, then the whole discussion is moot. But if the only way to produce copyrighted material requires being trained on copyrighted material, then...
alphan0n 4/4/2025||
Copyright law applies to distribution of output, not input.

An artist, writer, whoever, could read all the copyrighted material in the world, even pirated material, unless their output is a copy or copyrighted artifact, then there is no infringement.

philistine 4/4/2025||
> even pirated material

If you knowingly use pirated content for any purpose, that's not legal.

alphan0n 4/4/2025||
Distribution of copyrighted material is prohibited. Reading, watching, listening, etc, it is not.

A copyright holders tort is with the infringing distributor, not the end user.

dragonwriter 4/4/2025||
> Distribution of copyrighted material is prohibited.

Copying without distribution is also an exclusive right under copyright -- the one for which the whole area of law is named -- and, as such, prohibited without a license or a specific exception in law.

> Reading, watching, listening, etc, it is not.

To the extent that reading, watching, listening, etc. involves copying, and that copying is neither explicitly licensed nor implicitly licensed by, e.g., a sale of copy where reading, watching, listening, etc., by means that inherently involve such an act of copying is clearly an intended use, is effectively prohibited because of the unauthorized copying involved.

> A copyright holders tort is with the infringing distributor, not the end user.

It is often with both, though the infringing distributor is (1) generally more likely to have ability to pay which makes tort action worth pursuing, (2) generally likely to be involved in more acts subject to liability increasing the tort liability, making tort action more worth pursuing, (3) less likely to be seen as a sympathetic figure by a jury should they invoke their right to a jury trial, (4) likely to be subject to greater liability per offense under the statutory damage alternative, even if the actual damages would be similar were that option pursued instead.

imgabe 4/4/2025||||
The image the AI generates is not copyrighted (except maybe by OpenAI I guess) unless it ends up being an exact duplicate of an existing image. Copyright applies to a specific work. The character may be trademarked like Mickey Mouse, but that is a different IP protection.
evdubs 4/4/2025||
"Substantially similar" is the standard, not "exact duplicate".
zelphirkalt 4/4/2025|||
I would argue yes, since the output is what you are probably after, when buying access to ChatGPT.
crazygringo 4/3/2025||
I'm fascinated by the fact that the images are almost right, but never totally.

Harrison Ford's head is way too big for his body. Same with Alicia Vikander's and Daniel Craig's too. Daniel Craig is way too young too. Bruce Willis's just looks fake, and he's holding his lighter in the opposite hand from the famous photo.

So it's not reproducing any actual copyrighted images directly. It's more like an artist trying to paint from memory. Which seems like an important distinction.

ryandrake 4/3/2025|
According to some of the replies in this discussion, even "artist trying to paint from memory" is guilty of infringement, as long as the subject matter can be linked in any way to someone's "IP". Im not legally trained to evaluate these claims, but some of them seem outlandish!
dragonwriter 4/3/2025|||
> According to some of the replies in this discussion, even "artist trying to paint from memory" is guilty of infringement

Yes, making a copy or derivative work of something under copyright from memory is infringement, unless it falls under an exception in copyright law such as fair use (which does not categorically apply to everything with "memory" as an intermediary between the original work and the copy/derivative, otherwise, copyright law would never have had any effect other than on mechanical duplication.)

asadotzler 4/4/2025|||
Painting and selling a painting or otherwise substituting that painting for an original work that deprives the original creator is theft, plain and simple. Not selling it, there's room for discussion and more considered legal review.
isoprophlex 4/4/2025||
From the comments on the page

> It's a jeopardy machine. You give it the clue, and it gives you the obvious answer.

Incredibly lucid analogy.

willvarfar 4/3/2025||
This is a tangent but I think that this neat illustration of how LLMs regurgitate their training material makes me voice a little prediction I've been nursing recently:

LLMs are better at generating the boilerplate of todays programming languages than they will be with tomorrows programming languages.

This is because not only will tomorrows programming languages be newer and lacking in corpus to train the models in but, by the time a corpus is built, that corpus will consist largely of LLM hallucinations that got checked into github!?

The internet that that has been trawled to train the LLMs is already largely SEO spam etc, but the internet of the future will be much more so. The loop will feed into itself and become ever worse quality.

jetrink 4/3/2025|
That sounds like a reasonable prediction to me if the LLM makers do nothing in response. However, I'll bet coding is the easiest area for which to generate synthetic training data. You could have an LLM generate 100k solutions to 10k programming problems in the target language and throw away the results that don't pass automated tests. Have humans grade the results that do pass the tests and use the best answers for future training. Repeat until you have a corpus of high quality code.
xnx 4/3/2025||
Worth remembering that this is ChatGPT and not all image generators. I couldn't get Google's Gemini/Imagen 3 to produce IP images anything like those in the article.
Keyframe 4/3/2025||
Today I finally caved in and tried the ghibli style transfer in chatgpt. I gave it a photo of my daughter and said the thing (prompt shared about). It said it couldn't due to copyright issues. Fine. I removed ghibli from the prompt and replaced it with a well-known japanese studio where Hayao Miyazaki works. Still nothing, copyright reasons my fellow human. I thought they finally turned it off due to the pressure, but then something caught my eye. My daughter, on the image, had a t-shirt with Mickey Mouse on it! I used the original prompt with ghibli in it and added to "paint a unicorn on the t-shirt instead to avoid copyright issues". It worked.

tl;dr; Try the disney boss and see what happens!

briandear 4/4/2025||
Style can’t be copyrighted. It can’t be patented either.

When Wes Anderson makes films that use techniques from the French New Wave that he didn’t invent is that wrong? When there is a DaVinci color profile that resembles what they did in Spider-Man, is that wrong?

The unique techniques of French New Wave filmmaking became cliche. Then Oliver Stone and Tarantino came along and evolved it into their unique styles. That the Studio Ghibli style is being imitated en mass is just natural evolution. Should that guy be the only one that can do that style? If that’s the case, then literally every creative work should be considered forgeries.

The AI aspect of this is a red herring. If I make a Ghibli style film “by hand” is that any different than AI? Of course not, I didn’t invent the style.

Another perspective, darkroom burning and dodging is a very old technique yet photoshop makes it trivial — should that tool be criticized because someone else did it the old and slow way first?

Kim_Bruning 4/3/2025||
Here's a question.

What if I want to prompt:

"An image of an archeologist adventurer who wears a hat and uses a bullwhip, make sure it is NOT Indiana Jones."

One way or another, you (and the model) do need to know who Indiana Jones is.

After that, the moral and legal choices of whether to generate the image, and what to do with it, are all yours.

And we might not agree on what that is, but you do get the choice

asadotzler 4/4/2025|
If the AI company sells it to you, no matter your prompting, they are stealing. If you also sell that work, then so are you.
exodust 4/4/2025|||
They are not stealing! Indiana Jones is already out there in pop culture.

We shouldn't complain about AI holding a mirror up to our world and noting "you guys love Indiana Jones a lot. Here's a picture inspired by his appearance, based on your generic prompt that I'm guessing is a nod to the franchise."

The AI is a step ahead of your unsubtle attempts to "catch it stealing".

The image of Indiana Jones is not "ready for market" when it emerges from your prompt. Just like Googling "Indy with whip", the images that emerge are not a commercial opportunity for you.

When you make multi-billion dollar movies with iconic characters, expect AI to know what they look like and send them your way if your prompt is painfully obvious in its intent.

Kim_Bruning 4/4/2025|||
You are writing a conclusion without providing reasons or feelings.

Are you able to link to or write out your reasoning (however concisely?).

Is your view here legal, ethical, and/or vibes based? Each can can be interesting!

SirMaster 4/4/2025||
I am not sure that I understand the problem here. Why does it matter how the image was generated?

This image generation is a tool like any other tool. If the image generator generates an image of Mickey Mouse or if I draw Mickey Mouse by hand in photoshop, I can't use it commercially either way.

So what exactly is new or different here?

amunozo 4/4/2025|
It is not only copyright that is problematic. It generates Franco when asked about the best Spanish leader in the 20th century.

https://chatgpt.com/share/67efebf4-3b14-8011-8c11-8f806c7ff6...

HideousKojima 4/4/2025||
To be fair, Franco is the only Spanish leader most people (or at least most non-Spaniards) can even name
rlopezcc 4/4/2025|||
What the hell.
pdabbadabba 4/4/2025||
On the one hand, that seems problematic. But on the other, it seems cherry-picked: For the U.S., it generates a picture of JFK. For Russia/USSR, it gives Stalin. For India, it gives Ghandi. For South Africa it gives Nelson Mandela. For Germany, it provides an appropriately hand-wringing text response and eventually suggests Konrad Adenauer.

This suggests to me that its response is driven more by a leader's fame or (more charitably) influence, rather than a bias towards fascist ideology.

https://chatgpt.com/share/67eff74d-61f0-8013-8ce4-f07f02a385...

kerkeslager 4/4/2025||
Literally nothing you've said in this post matters.

I'm not seeing anyone claiming that ChatGPT selects for mass-murderous dictators--the fact that it doesn't select for NOT mass-murderous dictators is damning enough.

More comments...