Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask HN: DALL-E was trained on watermarked stock images?

news.ycombinator.com

Post reply on HN
Print view View original

221–230 of 233 posts

Re: Ask HN: DALL-E was trained on watermarked stock images?

#221

Post by still_grokking »

stubish wrote: post #217
still_grokking wrote: post #199

Earlier quoted context omitted.

> On the other hand, here's a specific prompt: "nerdy yellow duck reading a magical book full of spells" > Dall-E: https://i.imgur.com/FMKZ8zc.png How well it learned all the common prejudices! "nerdy" == wears glasses I'm applauding. I'm looking already forward to AGI based on the current approaches… It will lead us finally into a better world, for sure. /s

How do you visually show 'nerdy' without resorting to the glasses stereotype? Your prompt is specifically requesting a prejudiced image.

Sure. And the AI serves the expected stereotype.

Isn't that great? The world will become a better place with AI everywhere.

We need especially more AI in law enforcement, and such…

AI should make important decisions. Because it bears the same prejudices as humans. So it can replace humans just great. ;-)

Re: Ask HN: DALL-E was trained on watermarked stock images?

#222

Post by omnimus »

usehackernews wrote: post #219
omnimus wrote: post #210

Earlier quoted context omitted.

Indeed the genie is out. And while we will get some interesting AI uses ultimately this is degenerative tech. In the end we end up with less authentic, less unpredictible and less delightfull art. Instead we get the perfectly suited to us, predictible, mediocre stuff. I said it in comment above - yes people build on work of others but they also bring lots of their originality and intelect. Part of what people do is t…

I don’t believe we will lose the capability to create new original styles. If a prompter can describe the creation of a new style, the AI can create it. Using both iterations of image & text prompts, unique styles will come. The thinking is still done by the human prompter.

The value of the image is in the human prompter (in the overall concept) but the overall style - the aesthetic is stuck in the past. Its almost impossible to describe aesthetic in text without referencing examples of that aesthetic. Its the case of one image says more than thousand words. It has to be seen.

I am not sure finding new aesthetics is even the playingfield nowdays. Its probably not because we’ve been stuck for decades. Its more about cyclic trends of things forgotten. So who cares. But this will just solidify that even more. But yeah it has already happened and since the tech will be firmly in private hands everybody will be just exploited and pushed by it instead of it helping anyone.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#223

Post by ehsankia »

lioeters wrote: post #218
Terretta wrote: post #159

Earlier quoted context omitted.

I’ve been comparing Dall-E, MidJourney, and StableDiffusion. Goes to show how much training set and implementation choices matter. But in all cases, you have to think of the underlying labeled text-to-image sets as paint colors to mix, and prepare a palette accordingly. Still haven’t figured out how to get what I want, but to your point, one can get closer. - - - Not sure if this is why, but with OpenAI’s Dall-E, you…

> you have to think of the underlying labeled text-to-image sets as paint colors to mix, and prepare a palette accordingly. Very insightful tip on how to harness the "creativity" of Dall-E and the like. I see how the phrase "king of belgium" was too vague for Dall-E, so it didn't produce anything recognizable - but changing the words into known details, like "banker" and "salt and pepper hair", worked effectively to…

It's not that it's "vague", they intentionally throw off when you try to generate a photo of a named person. It's an intentional protection they put in. If you just do "king" it'll likely do fine, but if it's referring to a specific person it won't.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#224

Post by ehsankia »

kaetemi wrote: post #113
dmitriid wrote: post #70

Earlier quoted context omitted.

I find Midjourney to be biased towards an artistic representation (for some definition of artistic) When Dall-e is happy to produce children's scribbles or poor imitations.

Try 'poorly drawn ... by a 5 year old using crayons' in Midjourney.

That's honestly one of my favorite prompts. It's funny to think I use this state of the art AI to generate crayon drawings, but they look so great!

https://i.imgur.com/jKcNkat.png

Re: Ask HN: DALL-E was trained on watermarked stock images?

#225

Post by lioeters »

ehsankia wrote: post #223
lioeters wrote: post #218

Earlier quoted context omitted.

> you have to think of the underlying labeled text-to-image sets as paint colors to mix, and prepare a palette accordingly. Very insightful tip on how to harness the "creativity" of Dall-E and the like. I see how the phrase "king of belgium" was too vague for Dall-E, so it didn't produce anything recognizable - but changing the words into known details, like "banker" and "salt and pepper hair", worked effectively to…

It's not that it's "vague", they intentionally throw off when you try to generate a photo of a named person. It's an intentional protection they put in. If you just do "king" it'll likely do fine, but if it's referring to a specific person it won't.

Ah I see what you mean - "king of belgium" is a real person, so they put in some safe guards in DALL-E to prevent recognizable images for such queries. Makes sense.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#226

Post by gojomo »

wettico wrote: post #190
gojomo wrote: post #86

Earlier quoted context omitted.

Whether something is directly competing for the same business would have to be evidenced, and copyright doesn't mean protection from all possible competition - it's just one factor weighed. And fair use protects many commercial uses, too, depending on proportion/character-of-original/etc. But also, none of these images are direct, or even necessarily subtantial, "copies" of other images. The generator learned from ot…

"The generator learned from other images – the same as any human artist might." A lot of people seem to make this comparison, but I don't think it's fair. It's wrong. A computer is capable of ingesting/processing and "learning" from images at a rate no human can possibly come close to matching. To elaborate, it is not actually learning in the way we normally think of it, as its "brain" is completely different from a…

It's great that human artists learn from, & introduce into their work, influences other than just patterns seen in other works.

But it's also great that AI artists can learn from more examples in a few minutes than a human artist might see in lifetime.

To say that's "not actually learning in the way we normally think of it" is superficially true, but it doesn't mean it's "not actually learning", or necessarily any worse than typical learning. It's so new, & we barely understand fully how it works or what its limits are. It might be better in many relevant & valuable aspects!

Re: Ask HN: DALL-E was trained on watermarked stock images?

#227

Post by stubish »

still_grokking wrote: post #221
stubish wrote: post #217

Earlier quoted context omitted.

How do you visually show 'nerdy' without resorting to the glasses stereotype? Your prompt is specifically requesting a prejudiced image.

Sure. And the AI serves the expected stereotype. Isn't that great? The world will become a better place with AI everywhere. We need especially more AI in law enforcement, and such… AI should make important decisions. Because it bears the same prejudices as humans. So it can replace humans just great. ;-)

The AI is making no decision. The person entering the prompt made the decision to include the term. It is performing the same function as a pencil.

Now, if 'criminal' rendered as a black male 90% of the time rather than a crouched white male wearing a cheesy burglar mask and a sack over his shoulder, then I could see your point about perpetuating prejudice rather than stereotypes.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#228

Post by wettico »

gojomo wrote: post #226
wettico wrote: post #190

Earlier quoted context omitted.

"The generator learned from other images – the same as any human artist might." A lot of people seem to make this comparison, but I don't think it's fair. It's wrong. A computer is capable of ingesting/processing and "learning" from images at a rate no human can possibly come close to matching. To elaborate, it is not actually learning in the way we normally think of it, as its "brain" is completely different from a…

It's great that human artists learn from, & introduce into their work, influences other than just patterns seen in other works. But it's also great that AI artists can learn from more examples in a few minutes than a human artist might see in lifetime. To say that's "not actually learning in the way we normally think of it" is superficially true, but it doesn't mean it's "not actually learning", or necessarily any wo…

Fair, I don't know what it's actually doing. I just know you can't equate it with anything a human does, and the use of the word "learn" is misleading, or vastly oversimplifies what is happening, to the point that it allows for false analogies.

That said, my main objection to this technology is that:

- The AI's work is based on human artists' work

- Companies are then profiting off of the AI's work

- The companies are indirectly?/directly? profiting off of artists' work

- The companies do not get artists consent or compensate them in any way

- The companies are essentially stealing from artists

Companies should be forced to obtain the creator's consent when using art to train their models.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#229

Post by snickerbockers »

throwaway675309 wrote: post #91
snickerbockers wrote: post #55

Regardless of whether or not training an AI on stock images violates the license, there's a very real problem with that watermark being present, which is that it proves their AI is prone to copying large swaths of images from gettyimages unaltered, and that definitely is a license violation. This makes me think back to the controversy over github copilot; if these AIs are going to be trained on other peoples' IP then…

Just because it contains the text of the watermark does not mean that it's reproducing large swaths of the image - its doubtful even the most generous perceptive hash would retrieve any matching images in the Getty repository.

and yet the watermark is there. if it can't copy parts of the training images then how and why did it copy the watermark?

Re: Ask HN: DALL-E was trained on watermarked stock images?

#230

Post by sixothree »

dlg wrote: post #30

I am not a lawyer, but I've had to argue about copyright with several. In the United States, there are two bits of case law that are widely cited and relevant: In Kelly v. Arriba Soft Corp (9th), found that making thumbnails of images for use in a search engine was sufficiently "transformative" that it was ok. Another case, Perfect 10 (9th), found that thumbnails for image search and cached pages were also transforma…

> (2) creative nature of the work

Is AI even capable of having a creative nature. All that I see is re-use of source images.

Post reply on HN