Midjourney Showcase | Hacker News Reader

Midjourney Showcase

midjourney.com

39 pointselsewhen65 comments

"Want to browse millions of images like these [...]?"

No, I really don't. Is anyone working on new ways to manage the flood of machine generated content? I suppose it's foolhardy to try and organize whatever gets created, if I'm looking for something in particular I can just generate it on the fly.

I saw an article a few days ago that stated that AI is about to pass humanity in content created since the beginning of time.

AI will soon have surpassed us in the amount of generated images, texts, videos. It's truely a revolution and content availability will not be a problem of the future, just to filter out what you want to see.

Midjourney output is: - 70% depictions of conventionally attractive women, often with unconventional / unrealistic body proportions - 20% run of the mill fantasy book covers - 10% everything else (including some really cool and interesting images).

Btw, "conventional beauty" is effectively an underhanded slur rooted in the envy of the "are-nots" toward the "ares". "Unrealistic" functions similarly, as there are plenty of women with the body proportions depicted in the handful of images I managed to see quickly perusing this page, and so while they may be uncommon (made, shall we say, more conspicuous in an age where obesity has skyrocketed and slovenly dress is ubiquitous), they cannot be unrealistic as such women do exist and do, in fact, dwell among us as fellow human beings. (N.b., some of the women depicted are entirely average or vulgar, and the manner in which some women are 'depicted' is also deficient in taste.)

And as this is effectively an artistic depiction, it would not be unreasonable to expect some idealization (no wrinkles, no blemishes, etc), which is kind of the whole point, as art, at least fine art, as it is produced to contemplate and enjoy beauty (which is, pace the relativists, objective; subjective tastes may conform to greater or lesser degrees to the objective fact). Even when you get your photograph taken, don't you want your "good side" photographed, or the zit on your forehead concealed?

Eh?! Only on MJ women have boobs bigger than their head.

The outputs represent the intents of the users.

No shock with mainstream media resenting “conventional beauty” that people would lean towards it with AI tools.

The descriptions don't really have to indicate beauty standard though. Seems to be implied.

e.g.: "18th Century Regency Era Portrait painting of a woman with long hair, wearing jane austen cream floral"

"Asian woman, pastel colors, mystical, empowering, stand in your own power, infinite possibilities"

That sounds roughly like the same ratios for most digital art on the internet. It's also probably indicative of the types of people (kind of nerdy, probably) people who are making images with Midjourney. I'm more likely to use Midjourney to make an image than a trained artist is, but a trained artist's image will likely be higher quality, more creative, and more diverse from what we're seeing here. Clearly, I'm guessing a little bit, but I bet that's what's happening.

I don't know. Here's someone making vintage movie posters with Midjourney[0]. Most professional artists would kill to be able to produce work like this.

Also another account[1]. I don't see lack of quality and creativity here. Other than maybe the bland face syndrome which (apart from weird hands) seems to be a problem with AI.

Granted, one could say that what quality and creativity there is belongs to other artists (notwithstanding touch-up work in photoshop, etc) but it's still clearly there. We're way past the point where AI art is qualitatively substandard relative to professional art, it's just subject to Sturgeon's Law, where most of it is low effort.

[0]https://www.instagram.com/todd_skull/

[1]https://www.instagram.com/drawiaj/

Midjourney is a gateway drug to more advanced image generators.

Which generators are you referring to?

Someone else mentioned civitai.com, I gave it a look and was surprised at the diversity of models, most of which are derived from stable diffusion tho.

Stable Diffusion, likely.

This is an exaggeration, right?

Then again, I see your point; even if the actual numbers were half of that, it would be a lot.

It's tongue in cheek and of course I don't have comprehensive data, but I do follow AI art a lot (which I find interesting) and that's my general impression. I wouldn't even say that the numbers are that wildly exaggerated.

I have no idea how to put it in words and I am far from bashing it, I like it, admire it.. but.. it seems vaguely divergent styles, stil what it bugs me after seeing lots of these images, it seems "boring" (again i don't know how to phrase it).. if i see an image in the wild i can guess it is generated by midjourney.

>after seeing lots of these images, it seems "boring"

Midjourney very much has a style and like looking at the front page of ArtStation while the images are of a high “quality” in the technique sense, the artistry of them is pretty low.

Very rare to see an image that will surprise you on ArtStation, just a sea of orcs, elves, guns and spaceships in the same “painted” yet completely uninteresting style. Same for a MidJourney output, all very safe, all the same tricks to provide “impact” yet all utterly boring.

You can't generate anything "unsafe" with it. Its keyword filter won't let you.

"Safe" doesn't mean non-pornographic, it means not artistically interesting.

Pornography is not the only thing that's censored there. Any kind of nudity, violence, blood, some politicians and other things like that are banned too.

https://aipromptsguide.net/list-of-banned-words-in-midjourne...

I know exactly what you mean even though I basically disagree. While in some pictures, one can really "tell" by picking up on some unconscious "artifacts" or patterns. Uncanny valley stuff. On the other hand most are just too good, which I egocentrically assume is what you also mean. (or at least it makes sense from my perspective)

This leads me to believe that a lot of people may share your view and thus steer the upcoming popular visual styles into the opposite directions (the pendulum). Maybe more abstract / gritty / grunge / minimalist or messy. I've seen a similar development in music and it steered the popular audience to the stoner rock genre for a while, simply for being the opposite of the currently popular in multiple ways.

Good for Ilford, I bet. Film in the '20s is in a great place to lean into the trend.

Midjourney gives little control to artists and straight up ignores prompts. You can't customise it with different loras and checkpoints. There's no controlnet. It's tuned to "look cool" not to output what's actually in the prompt.

For example - if you prompt "low poly castle, isometric". The output wont't be low poly or isometric, but something that "looks cool".

This is why people are hyped about the new SD XL. It understands prompts much better than previous iterations of SD and Midjourney. And it can be customised and extended. Can't wait for the full release, Controlnet XL and other extensions.

>>if i see an image in the wild i can guess it is generated by midjourney

That's because of default settings, it's not a limitation of the model. I get Midjourney magazine for $4/mo; curated examples are quite amazing.

Are these human-created images less boring? https://dribbble.com/shots/popular

I like the cat on the doorstep. But that's 100% a style I see appropriated by stable diffusions.

More boring, only because dribble's popular page is UI mockups.

A better comparison is this vs artstation.

Would be fun if someone made an online game where you try to distinguish AI and human generated art. I'm confident I would do much better than chance, but I'd be curious to see if people can reliably get 90%+

There appears to not be that much diversity in facial structure, body types, clothing styles, color, sex gender identity, ethnicity, setting, perspective or culture in the human beings present on that page.

This leads me to question the AIs training biases and capabilities to generate truly representative art depicting the human experience with the tool.

It also represents author biases, of course.

The training sets that form the foundation of this iteration of AI reflect our collective culture.

People who complain about biases have only themselves to blame.

The more 'diverse' the people that get involved in our culture - getting their hands dirty and actually creating new things, instead of parasitizing off the efforts of others - the less bias there'll be. Simple as that.

It doesn't represent human experience, it represents the average submission to deviantart

Still wild to me that their tool only exists in Discord.

I know a lot of artists feel threatened with Midjourney/Stable Diffusion. I wonder if there would be less apprehension if these tools were built more like Photoshop instead of a command line? Augmenting photo creation/editing rather than looking like it replaces it all.

There already are SD plugins for Krita, Photoshop, GIMP and Blender.

Is the user interface still Discord?

Indeed. It sort of blows my mind that they cannot put a simple web UI up given all the resources they have at their disposal. To get midjourney, I need a midjourney account *and* a Discord account (I'm not a gamer).

There's no such thing as a simple web UI - discord is on all platforms desktop and mobile, they would have to duplicate a lot of work. It's a neat hack they recognize that generative AI is essentially conversational (or to go the other way, a conversation is essentially an iterative walk through possibility space).

EDIT: I don't know what you mean by having a discord account and a midjourney account, you log into billing by granting it access to your discord account - only one username/password is necessary.

Why would they have to recreate the wheel? A lot of very large companies use Gmail or Outlook for their emails, or Slack for their communications...

Getting real time communication at scale is real tough, especially when Midjourney is from far the largest Discord server https://www.statista.com/statistics/1327141/discord-top-serv...

> Getting real time communication at scale is real tough

Fair enough, but maybe I don't want a community. Maybe I want an application that works like Photoshop, Gimp, Inkscape, Figma, etc. Or maybe I just want a query response website, like a classic search engine, or how OpenAI works.

I think the refusal to put up a web app just seems lazy.

> It sort of blows my mind that they cannot put a simple web UI up given all the resources they have at their disposal

Of course they could if they wanted to.

They think it's good to have the whole community on their discord channel, so people can see what other people are doing to inspire new users.

I'm not on it myself, so I can't comment on whether it works.

Discord dropped the “for gamers” thing a while ago, and for example many open source projects use it to host their community and provide support

I don’t think that means they should only be available via Discord but not being “a gamer” isn’t a good reason to not use Discord on its own

can you elaborate on why creating 2 accounts is such a turn off? Don't get me wrong though, I agree that discord is a horrible UI for such a powerful tool.

How many accounts would be too many? Is 4 accounts too many? 6 accounts?

Some of us consider more than 1 too many.

Creating even one new account is usually more of a hassle than I am willing to deal with.

this is absurd. i have no idea why teens like discord so much, it's like irc with all the bad sides (messy, not stored on the open web, ephemeral, corporate censorship) and without the good sides (privacy, decentralization, etc)

> (privacy, decentralization, etc)

These are not universal values.

> (messy, not stored on the open web, ephemeral, corporate censorship)

Being stored on the open web to be indexed and archived is really the opposite of private. Obviously discord has access to everything but as far as the user is concerned its more private than anything else short of e2ee group chats (but what's the point of an e2ee chat with hundreds of strangers, any one of which can be a fed?)

And I don't know if you noticed but a lot of the youth prefers the censorship, or as they call it, taking out the trash.

Hosting a chatroom yourself is too much work. The same reason why nobody wants to run their own email server.

IP in the style of IP

There’s also an official Midjourney monthly paper magazine I love (https://mag.midjourney.com/). A selection of images and prompts. Genius blend of new and old.

Interesting technology but I'm not seeing anything there that makes me think any of this is "art" in any conventional sense of the word.

Lots of "snapshots" but very few photos, if that makes sense to anyone.

Seems like girls are the preferred ones to show the human beauty. Wonder why.

Probably because most of the users are men?

because we don't say a beautiful man, we say a handsome man

As a side note , pretty sure 90% of stable diffusion generated images is porn. If you go to civitai and browse for the top images, it's mostly that. I assume it's the case here as well.

AI-generated porn is a weird hobby with some very vocal supporters. Everyone else is pretty happy with real porn or rule34 or certain artists with Patreons.

People use Stable Diffusion because it's free. I don't really want to pay for a Midjourney subscription, but I'm happy to spend a few cents of electricity refining prompts and batch generating images to use as placeholder art.

Not sure about the exact percentage, but are there any other image generators that can be run locally and have no censorship?

I only know of stable diffusion. The biggest factor, IMO is the trained checkpoints, and with Civitai website for SD, there's really an unmatched amount of checkpoints and Loras , so it's your best bet.

I see lots of large breasted women. I'm guessing MJ's users skew young and male. There is nothing wrong with it, but it really kind of sticks out.

Perhaps an "/epicfail/recent" gallery is also needed. Show the images that are horribly wrong. No porn please.

So much of the media we consume has been focus-grouped, sanded down, moderated, de-fanged and otherwise diluted into bland corporate mush that it makes sense that we'd want to be able to skip the creative process altogether.

Generating images through software hooked up to statistical models trained on enormous amounts of other people's intellectual property is a cool trick, but it's not art.

A banana taped to a wall is considered art these days.

With duchamgpt-v2.3 you can generate as many images of banana and banana-esq items affixed to surfaces as you can imagine, or, rather, not imagine.

I guess the good stuff is private and not on Showcase - quite poor quality here.

Only top paid tier allows private generations. The rest are public. But they might as well be private, because they quickly get buried.

You mean the showcase is poor quality?

Can you link to a midjurney image with better quality?

> API seems to be down, please try again later.

Great show.

Reply on news.ycombinator.com