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

111–120 of 233 posts

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

#111

Post by yreg »

smlacy wrote: post #39
webwielder2 wrote: post #38

These are the absolute worst DALL-E images I've seen. Do people generally just share the amazing ones and most of the output is actually complete shite? Like Instagram presenting the top 1% of people's lives.

Prompt Engineering can help a lot but yes, you're basically right: People are generating many, many images and sharing only the best ones with the fewest artifacts. For simple prompts with little additional guidance, all the diffusion image generators I've seen/used will produce output about like what the author linked most of the time. There are always a few gems, and honing in via prompt engineering helps immensely…

I disagree, with proper promptcrafting you can expect far far better results than the one in the op, before any cherry picking. (see my comment sibling to yours)

With Dalle-2 I get a satisfying result in >50% of attempts and I'm a beginner.

With Midjourney the result almost always looks great, but often misses some part of what I wanted. I'd say Stable Diffusion is similar. The results are seldom crap, but it's difficult to bend it to produce unusual situations. And in SD it's difficult to keep the entire objects in the frame, but that's a different problem.

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

#112

Post by NoMoreBro »

webwielder2 wrote: post #38

These are the absolute worst DALL-E images I've seen. Do people generally just share the amazing ones and most of the output is actually complete shite? Like Instagram presenting the top 1% of people's lives.

There is always some cherry-picking, but prompt engineering is an art per se, you become better and better working on it. I just started this experiment https://www.instagram.com/unshushproject or without Instagram https://unshush.com and spent A LOT of hours and patience to become good at it. Now I'm very proud of my results and I'm working on doing better.

It's a bit risky to invest too much time because every generator is different and they change the underlying model frequently (see the beta of MidJourney yesterday), but if you do it for passion or curiosity there is no problem.

Now I'm experimenting with a local installation of Stable Diffusion (well, not really "local" because I have an old computer) and the prompt is only one of the things you can tweak. There are num_inference_steps, guidance_scale and other parameters.

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

#113

Post by kaetemi »

dmitriid wrote: post #70
ehsankia wrote: post #44

Earlier quoted context omitted.

Top 1% is a bit exaggerated, but there is definitely a lot of not good stuff. I find that Dall-E does especially poorly with underspecified prompts too, unlike something like Midjourney which can give visually pleasing photos for even the most abstract concepts. Dall-E tends to do better with concrete and specific prompts. Here's an example: Stressful Shapes Dall-E: https://i.imgur.com/JBkSh0y.png Midjourney: https:/…

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.

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

#114

Post by bryanrasmussen »

gojomo wrote: post #48
eslaught wrote: post #33

Earlier quoted context omitted.

Search engines don't create market harm for a work because they don't compete with it. In fact, they do the opposite: they advertise the work, making it more accessible and increasing exposure. These AI tools on the other hand seem to do the exact opposite. They can (or could, if they got good enough) absolutely compete with a work, and therefore seem like they create substantial market harm. The character of use als…

As I see it, 3 of the 4 tests are strongly in OpenAI's favor; the 'market effect' is mixed. (1) The use is highly transformative; (2) the images used were offered to the anonymous browsing public (with watermarks); (3) the end effect of training will only retain a tiny spectral distilled essence of any individual photo, or even a giant source corpus; (4) there's a potential risk of market competition from the ultimat…

aside from if it is not copyrighted the image, the Getty watermark usage probably might have a bunch of issues.

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

#115

Post by kaetemi »

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…

It's prone to copying things it has seen thousands of times, such as those watermarks. The content itself is unrelated.

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

#116

Post by neurostimulant »

webwielder2 wrote: post #38

These are the absolute worst DALL-E images I've seen. Do people generally just share the amazing ones and most of the output is actually complete shite? Like Instagram presenting the top 1% of people's lives.

Yes, "prompt engineering" in actually a thing. People shares various tips and tricks on the internet to engineer their prompts for best results. Lots of trial and errors required.

Example: https://news.ycombinator.com/item?id=32088718

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

#117

Post by naillo »

webwielder2 wrote: post #38

These are the absolute worst DALL-E images I've seen. Do people generally just share the amazing ones and most of the output is actually complete shite? Like Instagram presenting the top 1% of people's lives.

They can definitely be that bad quite frequently. I've actually been a lot happier with stable diffusion outputs lately (doesn't hurt that they're free too).

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

#118

Post by ale42 »

jokethrowaway wrote: post #78
musicale wrote: post #77

Earlier quoted context omitted.

> As soon as the rulemakers caught up, it became obvious that they didn't have a license to operate differently from everyone else, just because they're new and popular. This is well said. One of the primary advantages of these businesses was evading the regulation and taxation that their competitors were subject to.

Indeed, but this is the proof the market didn't want the regulation and taxation and wanted better technology. Shouldn't we listen to what people want over bureaucrats?

People want a lot of things, but people can be egoistic. What people want is not necessarily good for the society as a whole. That's why laws exist.

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

#119

Post by ale42 »

donkarma wrote: post #22
RcouF1uZ4gsC wrote: post #7

What is interesting is a human analogy. Say you were an artist who went to every art show and museum and studied all the art there. If you produced a work of art solely from memory that contained large portions of other people's copyrighted art, would that still fall under copyright/require licensing?

yeah except this artist wont go around painting watermarks

There's a big difference here. DALL-E (or any other similar tool) is not an artist, it is a commercial product that can produce something that might be considered art. The usage might be ok, but the product itself might not... if I paint using stolen pigments (not saying that DALL-E is alike to this), my painting would be legal, but the fact of having stolen the pigment is very illegal...

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

#120

Post by ale42 »

8note wrote: post #75
RcouF1uZ4gsC wrote: post #7

What is interesting is a human analogy. Say you were an artist who went to every art show and museum and studied all the art there. If you produced a work of art solely from memory that contained large portions of other people's copyrighted art, would that still fall under copyright/require licensing?

Alternatively, if you memorized some GPL code, can you write a copy of it and put in a proprietary licence?

Of course not! That's why people do black-box reimplementations without ever seeing the original source code.
Post reply on HN