Comment by ssalka | Hacker News Reader

Comment by ssalka | Hacker News Reader

Parent Full thread

I think the headline is overly broad, especially considering:

> As a matter of statutory law, the Copyright Act requires all work to be authored in the first instance by a human being. Dr. Thaler’s copyright registration application listed the Creativity Machine as the work’s sole author, even though the Creativity Machine is not a human being. As a result, the Copyright Office appropriately denied Dr. Thaler’s application.

It seems like Dr. Thaler's argument was just weak, since generative AI works often are authored in the first instance by a human being. For instance, any Midjourney or Stable Diffusion-generated image will be sourced from a prompt, which is typically written by a human. Anyone who has spent a little time trying to craft the perfect prompt knows there is a creative process therein that represents real work being done by a human. Similarly for img2img workflows, using a real photograph taken by a human. There, AI is only being used to transform a copyrightable input. Therefore such works – though certainly not all AI works – should be eligible for copyright, IMO.

Thaler seems to go out of his way to claim no human intervention and authorship by the AI - So yeah, that's a very specific ruling that has little to do with AI as a tool. It's really more about AI personhood.

What's potentially more of a problem is the mention of artists using Midjourney and denied copyright - and very much separate cases from Thaler.

>Thaler seems to go out of his way to claim no human intervention and authorship by the AI - So yeah, that's a very specific ruling that has little to do with AI as a tool. It's really more about AI personhood.

This, it was a poorly concocted scheme. People do stuff like this all the time, but even when they manage to confuse one branch of the government, the rest of the government isn't suddenly obligated to go along with it.

I don't feel that it was a misguided attempt to "get copyright". Wasn't the attempt specifically to get copyright attributed to the AI (rather than to Thaler)? So it was some grand scheme about AI personhood or business plan about selling software that would own its output. Who knows. Whatever. Not relevant to copyright for AI as a tool.

The copyright office has already ruled recently that prompts are not enough to gain copyright no matter how detailed or how many iterations.

Furthermore, the Copyright Office stated that prompts alone do not provide sufficient human control, as AI models do not consistently follow instructions in the prompts and often "fill in the gaps" left by prompts and "generate multiple different outputs"

Just zooming in on txt2img, an AI generated image is:

- The text prompt

- The negative prompt

- The model

- The model seed

- Any LoRAs selected

What about this is copyrightable? The specific text used in the prompt? This would mean I could copyright "man holding apple".

Maybe the summation of all of the above? But that would be akin to copyrighting a specific Adobe Photoshop workflow.

The prompt is a literary work independent of the system processing it. If the text is sufficiently elaborate, it is certainly copyright able. But the resulting image is still a different affair.

That sounds to me like a recipe. Recipes their interaction with copyright are well-established, legally speaking, in the United States.

Their interaction being they are in no way copyrightable because they are functional, not creative expressions. That's part of why every recipe has a dramatic story, so they can have a clear copyright case if copied wholesale.

Is that really the reason for these stories? I only know about them from memes, and looked it up when I first read about it. In my language this trend hasn't caught on yet, thankfully. I always chalked it up to cultural differences (and judged Americans a little bit for it tbh, since the idea of integrating a story into a recipe sounds rather insane).

>Is that really the reason for these stories?

No, it's more about ads and SEO and the fact that a lot of people like the touchy feely aspect of the stories and that causes them to engage with the site for longer, leaving their own comments and returning to the site and such.

It is (or used to be) for SEO. Google Search loves walls of "relevant" text and will punish recipe sites that get right to the point.

It’s also to add vertical height so you see more ads on your way to the recipe.

It's more like a set of numbers since the prompt is just an interface that gets tokenized.

It would be like saying a DJ's equalizer settings are copyrightable.

You just stated how an AI generated image should be copyrightable. You should be able to own the copyright to all the configuration settings. If those settings then can be transformed to a 100% deterministic image (true, since you provide the seed) then I don't see how this is different than developing a photo negative film and transferring it to paper.

> This would mean I could copyright "man holding apple".

I think this is true today. You can have copyright on this phrase, just consider if it were the title of a song or poem.

> I think this is true today. You can have copyright on this phrase, just consider if it were the title of a song or poem.

That is not true today. You don't get a copyright on a phrase in particular if it was the title of a poem or song. For example:

"There's something in the way she moves" by James Taylor[1]

and "Something" by the Beatles[2] which starts with the same line.

James has the copyright over his song called "There's something in the way she moves" [3] and George Harrison's estate has the copyright over the one he wrote with the same title even though he probably copied it from James Taylor.

[1] https://youtu.be/p0FJUVo-BaM?si=fGR-TOim_8FS8rkO

[2] https://youtu.be/UelDrZ1aFeY?si=UG8c-cgfpgyH3I9n

[3] Which was the first one fwiw. He thinks because he signed to "Apple Records" (the Beatles' label) they heard his one when he recorded it before it was released and that maybe gave George Harrison the idea for the line.

I'm confused, are they both entitled to their copyright? (which I think should be the case)

They have copyright oveer their respective songs, not the phrase, so it disproves your statement that you can have copyright over a phase. It can be a trademark but not be copyrighted

If I write a program to generate text of random words, that output can't be copyrighted -- but the program itself is.

By the same token, the prompt is copyrighted - but not the output it generates.

Personally I'd like to see whether img2img works are copyrightable. My understanding is that copyright applies to the human-generated parts of an image. So e.g. In the case of a comic where the art is AI but the caption is human, the label but not the art is copyrightable.

How does that apply when we transform a copyrighted image? Is the resulting work covered by the copyright of the original? If so, can I create a bad sketch drawing, transform it with img2img, and get the result as copyrighted? If not, is there a specific denoising threshold at which copyright isn't applied?

Do you think ordering your burger medium rare is also human authorship?

The headline on Reuters seems to be more accurate (maybe it was changed after the article was posted here?). Unfortunately I can only got a glimpse of it before their overly-aggressive ad-blocker-blocker asserts itself (I’m fine with Reuters not wanting to serve me, since I block their ads, but their anti-adblocker system totally hijacks mobile safari).

You can click on "continue without supporting us".

I don’t see that option.

Anyone who has tried prompting AI to create an image should know it's not "trivial". It takes skill to get a good image, and the prompt itself is human creativity. The idea that the work produced is not from a human is insane. The model is just a tool like a camera.

Reply on news.ycombinator.com