Google Debuts Imagen 2 With Text and Logo Generation

Google's making the second generation of Imagen, its AI model that can create and edit images given a text prompt, more widely available -- at least to Google Cloud customers using Vertex AI who've been approved for access. From a report: But the company isn't disclosing which data it used to train the new model -- nor introducing a way for creators who might've inadvertently contributed to the data set to opt out or apply for compensation.

Called Imagen 2, Google's enhanced model -- which was quietly launched in preview at the tech giant's I/O conference in May -- was developed using technology from Google DeepMind, Google's flagship AI lab. Compared to the first-gen Imagen, it's "significantly" improved in terms of image quality, Google claims (the company bizarrely refused to share image samples prior to this morning), and introduces new capabilities including the ability to render text and logos. "If you want to create images with a text overlay -- for example, advertising -- you can do that," Google Cloud CEO Thomas Kurian said during a press briefing on Tuesday.

Need case law or legislation soon

[ranton]

But the company isn't disclosing which data it used to train the new model -- nor introducing a way for creators who might've inadvertently contributed to the data set to opt out or apply for compensation.

We really need some case law or legislation to clear up whether or not AI creators need permission to use public data, copyrighted or not, to train AI models. I personally don't think permission should required to use copyrighted data to train an AI, just like a human can read copyrighted books as part of their process of learning to write better themselves. But I know there is disagreement here, and we really need some case law soon to clear this up.

Re:

[Rei]

We really need some case law or legislation to clear up whether or not AI creators need permission to use public data, copyrighted or not, to train AI models.

The law is already abundantly clear on this (you do not need permission to automated-process copyrighted data to create new products and services, and that copyright only applies to non-transformative recreation of works, not styles or generalities), which is why basically Google's entire business model isn't illegal and why people who keep suing AI co

Re:

[serviscope_minor]

The law is already abundantly clear on this (you do not need permission to automated-process copyrighted data to create new products and services, and that copyright only applies to non-transformative recreation of works, not styles or generalities)

Is it though? The legal theory behind how EULAs work is as follows: The act of loading software into RAM constitutes a copy. By default you do not have the right to make that copy of the software you just paid for. You can be granted that right if you agree to th

Re:

[tlhIngan]

Is it though? The legal theory behind how EULAs work is as follows: The act of loading software into RAM constitutes a copy. By default you do not have the right to make that copy of the software you just paid for. You can be granted that right if you agree to the terms of a license, otherwise, no software for you.

Actually, you do have that right even with "all rights reserved" software - the US Copyright office has long recognized that loading a copy into RAM is perfectly legal.

But then we have open source

Re:

[backslashdot]

So if I see a particular artistic style say at a museum or on a show .. if I learn from that style I have to pay the original artist? Note I am not talking about producing an exact or near copy of the artwork itself. For example if I see a particular artist likes to use glitter in the background, and do that in my own artwork .. do I have to pay that artist?

“Inadvertently Contributed"

[WDot]

I suspect the attack on generative AI wrt copyright is not completely about copyright infringement. Many of these pieces of art that are trained on are freely available on different websites or social media platforms, with no paywall. Oftentimes they are shared and reposted, and the only “bad form” in many cases is explicitly cropping out the artist’s name or website before reposting. However, many of these artists live on commissions, and their public portfolio is essentially a free sampl

Re:

[dvice]

Amount of required images depend on the AI as you can pre-train AI so that it requires less images to train a new skill. And you can easily create millions of images simply by shooting a video of something and converting each frame into an image. This is especially handy when creating training data e.g. for hands, because you can get shots from different angles and with different gestures and holding different objects. Big companies like Google would have no problem creating tons of photographic images to c

Re:

[dvice]

> have they made recreating images and adjusting existing ones easy

Yes. You can give an existing image for the AI and ask it to modify it. E.g. "make this image look like original sketch that was done before drawing the actual result". Or "add some trees to the background" or "use the pattern in this image I give you on the sofa". Or you can drag the eyes with your mouse to make them more open or more closed.

This is not the best example of what AI can do, it is just one example. You draw a rectangle to e

Better source with actual images

[dvice]

Better source with actual images:
https://deepmind.google/techno... [deepmind.google]

Choose a better link

[DrXym]

How about linking to Imagegen [withgoogle.com], instead of some shitty tech website where all the links point inwards at other articles from the same website?

inadvertently contributed

[skinfaxi]

"creators who might've inadvertently contributed to the data set" is a funny way to say "creators whose work has been ripped off for profit."

Could I characterize it as "tellers whose cash may have inadvertently contributed to my wallet" after robbing a bank?

Of course creators need to be protected and compensated! If AI keeps stealing everything people make public, what do you think creators will do? They won't share new work to be stolen, which makes everyone poorer.