Am I the only one tired of seeing DALL·E /Stable Diffusion posts?

It's a Major Ongoing Topic (MOT) so we apply the SNI test: if an article contains Significant New Information, then it's probably ok, otherwise we'll tend to downweight it.

We learned this approach the hard way after HN got inundated by topic avalanches in the past (e.g. Snowden in 2013). It works fairly well, but it requires us to see the posts, so if you notice a follow-up submission that isn't singing particularly well for its supper, we'd appreciate a heads-up at hn@ycombinator.com so we can downweight it.

Past explanations, if anyone wants more, can be found by scrolling back through these links:

'significant new information' is key when there's a major ongoing topic - https://hn.algolia.com/?dateRange=all&page=0&prefix=false&so...

we tend to downweight follow-ups unless they're particularly good - https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

repetition is bad for curiosity - https://hn.algolia.com/?dateRange=all&page=0&prefix=false&so...

diffs are what's interesting - https://hn.algolia.com/?dateRange=all&page=0&prefix=false&so...

I'm not tired of it. I think it's the most exciting technology at the moment. IMO, ML/AI is only going to get crazier in the coming years. We're in a period of intense and rapid development that probably won't end soon. Personally I'd like to see more products than research papers.

I've played with them a bit. I understand at least a lot of the science behind it, and the tech is interesting, but what people seem to want to do with it is kind of boring to me. I've been working with generative art for a long time, but for some reason this doesn't really move me the way that far less technically impressive stuff does.

I'm especially bored with the "Will AI replace artists?" breathlessness. It's tedious. Of course not. These tools are each like an artist in and of themselves. Did Picasso replace Rembrandt? Did Warhol replace Picasso? Nah. No AI will create what I'd create, or what Takashi Murakami would create, or any given high school kid with a paintbrush.

If all you think art is is being able to complete a commission, then sure, you'd probably see this as replacing what you think the purpose of an artist is. But you'd be wrong, and wrong in such a boring way that it's not worth debating.

Even if these AIs started generating their own ideas, it would be amazing and fascinating and worth celebrating, but it's not replacing anything. Just adding something new and cool into the world.

I mean HN always picks up on the latest things and runs with it while it's in the zeitgeist. Just let it pass and let people advance it. It's technology. It's neat.

Think about everything else that was a thing everyone posted about and frame it the same way. "what are other use cases are there for the greater collective good?" And you'll see how silly this sounds

Not tired, but not impressed either. Looks cool, but I think it gets pretty boring after a couple of usages. Highly overrated and hyped if you ask me.

This branch of research and technology is going to have immense impact.

It is basically visualization of words and expressions, and I expect we will see it woven into many parts of our daily life in the near future.

It will not put artists out of business, although it may significantly impact stock photo services. Everywhere you look - any place that has a photo or illustration - may soon be using generated images. I even expect meta versions which can take large amounts of text (such as an entire paragraph or essay) and generate images which seem to fit the topic or feel.

This tech will give non-artists the ability to express and visualize ideas (for fun or profit), and it will be a boon to professional content creators who labor over image collection and selection.

Heck, at some point the images will be generated on the fly by our browsers for content or even ads.

Generative art posts are being posted from time to time on HN for years. (eg. There are many "This X does not exist" posts)

But recently it is booming. The reason is not related to DALL·E, but Stable Diffusion.

Stable Diffusion is 1) open sourced 2) the model is small and easy enough to run on people's computer. As a result, it can embed in a lot of tools and services, so people are getting excited and start building ecosystem and business around it. When people can create art without effort, how to streamline the creative process and dealing with giant amount of arts are going to be the problems that worth solving.

As for use cases, I can see it either replacing a lot of artists or being one of the most important tool for artists, or both. Also, it might flood the Internet with its creations in the following years so be prepared.

Eh. An overhyped technology where I'm thinking "oh that's neat" but meh, it's a trend I'm not interested in, one gets used to that kind of thing. It doesn't bother me until it starts having significant negative effects (like crypto hype).

No. It reminds me a bit of when Snapchat launched and suddenly everyone was posting selfies with terrible filters.

That said, the excitement is understandable. AI/ML have been hyped a lot and there have also been a lot of unresolvable ethical questions. So there's a certain sense of relief at something objectively cool and fun, as well as recapturing the 'wow' factor of what's possible with computers. Often the latter is reserved for a game or video demonstration of what can be achieved with skill + a great toolchain + new hardware (or newly affordable or tiny hardware). It's not so often that it applies to tools, or that new tools so dramatically lower the skill barriers to creating high quality work. Since humans are so visually driven and graphics have always been the fancy chocolate of computer technology, such impressive new tools hit at a visceral level for many people.

I think the biggest (but less obvious) use case is that this in combination with other tools like GPT-3 is suddenly very close to AGI. Up to now AI in computer vision has been about recognizing objects, entities, or behaviors for narrow tasks like driving or surveillance. Suddenly language models, scene decomposition, and visual feedback are good enough that we're close to having real conversations with computers: you can show your computer a picture and talk about how it has people and animals, what clothes the people are wearing, whether they're on a boat or in a field, what they're doing and so on. This is not that far off from sitting with a child and looking at a picture book together.

It's a developing situation, like a war or a natural disaster.

I don't mind them on HN because HN doesn't support image previews and I generally don't have to click in to articles. (Sometimes I'll read a few of the comments on HN, which again fortunately doesn't support image embeds.)

I mind them just about everywhere else. I have a hard time to describe it and I seem to be somewhat alone in it, but the images coming out of these algorithms generally tend to trigger some revulsion reactions in me. It's possibly just an uncanny valley relative, but sometimes that revulsion hits notes of true, disgusting horror. The images tend to smell too much like nightmares for my comfort. That simile is about the best I can explain it, I think, they all smell too much like nightmares to me.

I've dropped/blocked people and bots on social media for posting lots of generated images without CWs. I've threatened to drop more. I hate how much Facebook's algorithm tends to surface posts with photos over other content in general, but especially in this moment where many of those photos are these generated things that give me the creeps. There's no good way to set a filter to never show me a DALL-E/Midjourney/Stable Diffusion image ever again.

TL;DR: I don't mind conversing about these algorithms in HN, but please stop posting the images themselves.

I am, after the fiasco that was Github Copilot. It seemed like every post was filled with people dooming over how Copilot was going to make developers obsolete. Then I used Copilot, and I can't see how anyone could possibly think that would be the case. Maybe if all you do is write sort algorithms day in and day out, but otherwise, I think our jobs are quite secure, and I think that will be the case with SD / DallE too (especially after having used Dall-E 2).

I am hoping to soon see how we can generate a coherent set of characters in different artistic styles and perhaps build a graphic novel / comic strip using prompts.

A slightly higher order application of the DALL-E/SD/MidJourney primitives.

More like annoyed, because it is not something I can easily get into. But hey, still better than all those crypto/web3 posts.

I have always ignored them as I did not know what DALL-E was, but last night got reading the multiple stable diffusion posts and liked what I read. Interesting stuff.

Now after having read, I think it is good to have one odd post here and there as there are always some people not in the know, but I agree we are seeing too many of those.

We've just scratched the surface of the implications. I just read another story on HN about the difficulty Twitter has in identifying child sexual exploitation images, and how they have so far passed on a multi-billion dollar business opportunity because of it.

Generative art seems to make such identification much more difficult. Such tech can hardly be stopped from helping to image the full range of human fantasy. CSE and every other form of crime and taboo will be included. The volume of such media will increase by orders of magnitude. Whether they are or should be legal is a fascinating topic, but it matters little to the advertisers that bankroll business models like Twitter's.

I can't even guess if this results in more or less strictness around such fantasy content.

Well, I'm certainly tired of hearing all the bitching about it and other cutting edge technologies.

One of the most interesting developments in computation in my lifetime? Nah, let’s have the millionth article about privacy, crypto or regulation instead.

Saying that this is uninteresting because art does not advance the collective good is a pretty narrow view of humanity.

This too will pass.

I've been generating thousands of images with Stable Diffusion and my overall take is this is super primitive tech and it's unclear if/when there will be major breakthroughs in terms of "mainstream" use cases or general impact on society. I think it could revolutionize gaming or marketing if it gets to the point where any meaningful prompt instantly leads to mind-blowing images. Currently you have to keep laboring against its tendency to produce grotesque, mutilated, dystopian, hell-like imagery.

I have been thinking the same. These are about art and impressive as they seem, it’s curious why so many non-artists are talking about it so much. I would be far more interested in a writing assistant that functions similarly to GitHub CoPilot, but works in emails and docs of various kinds. This would actually impact my day to day. I have yet to see a near-universal plugin that helps with writing. Anyone know of something that’s free or cheap (similar to copilot $10/mo would be fine)

Not sure what "every other posts" you are talking about. The current front page one is the only mention of Dall-E I have seen on this site in weeks.

AI techs is really good at "artistic" domains where the product is to make people feel. Can't be used where u want 100% correct output.

I do not appreciate the structure of the question. Half the front page has not been taken over by this question; moreover, no one is trotting out standard morality (collective good) frameworks, eg an act is good/moral because of its final consequences, who does it, or its intention. Those are the main lines. And then there are many many gambits, to re(mis)use chess terminology.

So: what are you asking, exactly?

If you weren't they wouldn't be upvoted and on the front page.

Per the guidelines[1], you can flag a post if you'd like.

Personally, the generative art part is pretty exciting to me. It's a good way to come up with new ideas. I don't see that usefulness going away.

1. https://news.ycombinator.com/newsguidelines.html

Better than political/web3/VC worship posts

It's obvious. What was once a mysterious domain can now be constructed with machines. It brings the concept of "art" closer and closer to the "axiomatic theory of art". We are understanding more and more what we are from a logical standpoint.

Actually this might not be for the greater collective good. It might be bad.

I'm tired of people acting like it just came into being out of nowhere. I've known about the training of the model for months, and about the predecessor Latent-Diffusion for a long time. Then there was DALLE-MINI/DALLE-MEGA. 4chan was all over SD and the leaked model before Hackernews was raving about it.

> what are other use cases are there for the greater collective good?

It could radically change how humans produce and perceive art.

I just wish there were more technical discussions around prompt engineering, so far most of the articles focus on particularly good examples authors got out of DALL-E, MJ, SD.

Nope. I said it too, earlier on. But no-one ever pays any attention to my words of wisdom:

https://news.ycombinator.com/item?id=32659058#32659107

Not yet, but likely will once they're in mass use and add to existing internet/content pollution. Not saying early adopter content os any better, just wary of the quality to volume of whatever will be produced.

You aren’t alone.

> Am I the only one tired of seeing DALL·E /Stable Diffusion posts?

No, there's quite a few of us. https://imgur.com/a/9gXjJ61

I'm more tired about calling it "art". It's about generating digital images. Results are mostly generative. Like artstation trending images, but with lack of details.

Plot twist: These posts are all AI generated. Including this one.

Just ignore them then.

The progress demonstrated in these models, particularly SD img2img is quite spectacular, and fully warrants the hype, IMHO.

Is this the thing with the images of melted people?

I find corruption of the human form, in general, in lousy taste and wish people would cut that out.

Just my $0.02

Not tired no.. its moving so fast I'm slightly scared of its rate of progression.

So I'm watching it in amazed fear.

It'd be nice to have filter words in our profiles that would excise matching posts from the front page.

META: Isn't this one?

I'm not tired of it, as we often have "deluges" of posts about current events.

It'll die down.

No offense intended, but I'm also tired of personal "Ask HN" posts.

It is kind of similar to cryptocurrencies in that it has no real use case IMO.

No.

Good native ad team though.

Upvote other content then? I think this is an artifact of the collaborative filtering model

Yes.