There is circumstantial evidence out there that 4o image manipulation isn't done within the 4o image generator in one shot but is a workflow done by an agentic system. Meaning this, user inputs prompt "create an image with no elephants in the room" > prompt goes to an llm which preprocesses the human prompt > outputs a a prompt that it knows works withing this image generator well > create an image of a room > and that llm processed prompt is sent to the image generator. Same happens with edits but a lot more complicated, meaning function calling tools are involved with many layers of edits being done behind the scenes. Try it yourself, take an image, send it it, and have the 4o edit it for you in some way, then ask it to edit again, and again, and so on. you will notice noticeable sepia filter being applied every edit, and the image ends up more and more sepia toned with more edits. This is because in the workflow that is one of the steps that is naively applied without consideration of multi edit possibility. If this was a one shot solution where editing is done within 4o image model by itself, the sepia problem wouldn't be there.
No elephants: Breakthroughs in image generation
141–150 of 373 posts
Re: No elephants: Breakthroughs in image generation
#142But the annotations are still wrong,
https://substackcdn.com/image/fetch/w_1456,c_limit,f_webp,q_...
(nice URL btw)
The room, the door, the ceiling are all of a scale to fit many sizes of elephants.
Re: No elephants: Breakthroughs in image generation
#143Earlier quoted context omitted.
Right the difference is that it’s a large company looking at it then copying it and reselling it without credit, which basically everyone would understand as bad without the indirection of a model. Edit: the key words here are “company” and “reselling”
But it’s not copying and reselling — it’s imitation. Copying is controlled by copyrights. And imitation isn’t controlled by anything. As for a company: a company is just a group of people acting together.
#1, it’s extremely easy to coerce direct copies out of models that artists could be pursued for infringement if they drew, but companies reselling said copyrighted artwork face no penalty
#2, yes, it’s a group of people who came together to build an algorithm that learns to extract features learned from images made by other people in order to generate images somewhere between these images in a high dimensional space. They sell these images and give no credit or cash to the images being “interpolated” between. Notice this doesn’t extend to open source, it’s the commercial aspect that represents theft.
The reality is that laws are meant to be interpreted not by their letter but their spirit. The AI can’t exist without the hard work its trained on, and the outputs often resemble the inputs in a manner that approaches copying, so selling those outputs without compensation for the artists in the training set should be illegal. It won’t be, but it should.
Re: No elephants: Breakthroughs in image generation
#144Earlier quoted context omitted.
You don't see a massive difference in the shear number of images that the AI can look at and the speed at which it can imitate it as a fundamental difference between AI and a human copying works or styles? For a human it took a lot of practice and a lot of time and effort. But now it takes practically no time or effort at all.
It took a truly colossal amount of human time and effort to build AI systems. It takes significant amount of energy to run those AI systems. I don’t see any meaningful difference at all between the system of a human, a computer and a corpus of images producing new images, and the system of a human, a paintbrush, an easel, a canvas and a corpus of images producing new images. Emphasis on the new — copying is still cop…
>It took a truly colossal amount of human time and effort to build AI systems. It takes significant amount of energy to run those AI systems.
Those people and effort aren't at all tied to the people who are making and using the art.
In the past every individual person would have to individually study art and some style and practice for years of their life to be able to replicate it really well. And for each piece of artwork it could take them days to make 1 single piece.
I would argue that this is why it wasn't really problematic to copy someone's work or style. Because the individual time and effort per person to even do that was so high.
But now that time and effort for an individual is next to nothing.
Re: No elephants: Breakthroughs in image generation
#145Earlier quoted context omitted.
In music, someone can sing the same style as another, but if they imitate it to the point that there is brand confusion, where the consumer believes the product came from X when it actually came from Y, that's clearly crossing the line.
Is that actually crossing a line? I'm sure some consumers have thought that rocket league was associated with FIFA, or that studio Ghibli movies were made by Disney. But these aren't widespread issues because we have a robust system of trademarks that draws a clear line: you can't use trademarked names or iconography in a way that causes confusion. But if some people heading Olivia Rodrigo's "good 4 you" think they'r…
vanilla ice vs queen case doesn't support your claim.
Re: No elephants: Breakthroughs in image generation
#146Earlier quoted context omitted.
You don't see a massive difference in the shear number of images that the AI can look at and the speed at which it can imitate it as a fundamental difference between AI and a human copying works or styles? For a human it took a lot of practice and a lot of time and effort. But now it takes practically no time or effort at all.
Well yeah, but copyright infringement isn't a function of how quickly you can view and create works. Copyright is meant to secure distribution of works you create. It's not a tool to stop people from creating art because it looks like your art. That has been a thing for centuries, we even categorize art by it's style. Imagine anime was had to adhere to a copyright interpretation of "it's my style!".
Current copyright yes.
But do you not for a second think that the current way the laws and rules are set are because of how hard and time consuming it was to replicate work?
Just because "that's how it's always been" doesn't mean it's acceptable to keep it that way when the means to perform the action have so drastically changed.
Re: No elephants: Breakthroughs in image generation
#147This is a before/after moment for image generation. A simple example is the background images on a ton of (mediocre) music youtube channels. They almost all use AI generated images that are full of nonsense the closer you look. Jazz channels will feature coffee shops with garbled text on the menu and furniture blending together. I bet all of that disappears over the next few months. On another note, and perhaps other…
I love using LLMs to generate pictures. I'd call myself rather creative, but absolutely useless in any artistic craft. Now, I can just describe any image I can imagine and get 90% accurate results, which is good enough for the presentations I hold, online pet projects (created a squirrel-themed online math-learning game for which I previously would have needed a designer to create squirrel highschool themed imagery)…
My problem with finding enjoyment in this is the same problem I have when using cheat codes in games: the doing part is the fun part, getting to the end or just permutations of the end gets really boring.
Re: No elephants: Breakthroughs in image generation
#148Earlier quoted context omitted.
You don't see a massive difference in the shear number of images that the AI can look at and the speed at which it can imitate it as a fundamental difference between AI and a human copying works or styles? For a human it took a lot of practice and a lot of time and effort. But now it takes practically no time or effort at all.
It took a truly colossal amount of human time and effort to build AI systems. It takes significant amount of energy to run those AI systems. I don’t see any meaningful difference at all between the system of a human, a computer and a corpus of images producing new images, and the system of a human, a paintbrush, an easel, a canvas and a corpus of images producing new images. Emphasis on the new — copying is still cop…
You don't see a difference between a person spending years learning techniques to create art by hand, and spending months or years studying and practicing some famous artists style, and then spending days manually crafting drawing a single piece of artwork in the style and quality of the originals.
The difference between that, and a person just entering a prompt to create some drawing in some style.
The model looked at orders of magnitude more examples of artwork than a single human could look at and study in a lifetime.
To me there is a clear difference here.
I am merely saying that perhaps the rules should change due to the drastic change in time and effort required to do the work.
Re: No elephants: Breakthroughs in image generation
#149Earlier quoted context omitted.
Couldn’t Ghibli zeitgeist moment lead to them making out hugely with a new release or just a cinema screening of Totoro right now?
That is "artists should be grateful to work for exposure" on a grand scale.
Except they didn’t do any work for the exposure. If a marketing agency had come up and executed the Ghiblify everything model as a PR stunt we would call it the most genius creative campaign of the decade
Re: No elephants: Breakthroughs in image generation
#150This is a before/after moment for image generation. A simple example is the background images on a ton of (mediocre) music youtube channels. They almost all use AI generated images that are full of nonsense the closer you look. Jazz channels will feature coffee shops with garbled text on the menu and furniture blending together. I bet all of that disappears over the next few months. On another note, and perhaps other…
I love using LLMs to generate pictures. I'd call myself rather creative, but absolutely useless in any artistic craft. Now, I can just describe any image I can imagine and get 90% accurate results, which is good enough for the presentations I hold, online pet projects (created a squirrel-themed online math-learning game for which I previously would have needed a designer to create squirrel highschool themed imagery)…
>I love using LLMs to generate pictures. I'd call myself rather creative, but absolutely useless in any artistic craft. Now, I can just describe any image I can imagine and get 90% accurate results
May I ask what you use? I'm not yet even a paid subscriber to any of the models, because my company offer a corporate internal subscription chatbot and code integration that works well enough for what I've been doing so far but has no image generation.
I have tried image generation on the free tier but run out of free use before I get anyway pleasing.
What do you pay for?