Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Scoops Novel]

I'm somewhat obligated to make this so you can all get rich investing in printing companies - who's the dodgy purveyor now, eh...

DALL-E 2, in the last few days, has "Cracked the Code on True AI-Generated Art." Put in a sentence - get the art. Literally that simple.

So, seeing as this is the biggest thing since writing, you can expect the world to change a little bit.

[Loud Whispers]

I'm not sold; art is already pretty diluted of meaning and an AI throwing more noise into the mix won't save art from itself

[Scoops Novel]

Thinking ridiculously smaller then the implications of the technology.

It's the biggest thing since writing. We only write because it's faster then drawing...

[None]

And this produces computationally-expensive nonsense. Writing will still be faster.

Also, it's worse at drawing hands than most artists are, ehehe.

[Scoops Novel]

Exactly what you want within minutes? Where are you getting nonsense from?

It's not even going to be expensive, judging by current prices for this kind of stuff.

With stuff like cross-language communication this will become a big deal IRL. For anything online where you used to use an emoticon, expect a illustration. For anything conceptually dense, expect illustrations. It's just going to be everywhere man.

[None]

Minutes, precisely. This would take minutes to generate at least using very powerful hardware (hence 'computationally expensive') and a massive body of reference material to make images from. I dunno if you have ten minutes for a coffee while the magic picture maker renders 'where is the toilet,' but cross-language communication is already a solved problem and this one's not worth having an existential panic about either.

Also, 'exactly what you want' won't be any kind of abstract, given the example image provided in the article you linked barely coheres- the horse is split in half and the 'astronaut' is very loosely placed on top of it with stupid sausage limbs and no meaningful texture whatsoever.

[Scoops Novel]

Very powerful hardware in the cloud. The images are only getting generated faster and faster - minutes is on the slow end.

It doesn't have to be perfect, it just has to be so good it always gets the point across.

Where before you would have to have memorised some lines for snappy communication and brought a video for anything really fancy and quick, now you can just show up with a drawing. And there's so many layers of communication you can do in a comic that you can't do in google translate. Simply being able to pick the style is a massive gain.

[FantasticDorf]

A discord discussion i was sitting in yesterday was pouring over the images being put out by DALL-E 2, mainly to admire but to also talk about what they could do with such a tool personally since they're into DND, SS13, etc where representation of the art, and production of art itself is quite important to visualizing the setting.

The waiting list for DALL-E 2 has quite a few twitter etc influencers right now from those who get approved, looking through it there's a childs book writer, furries*, & one of the devteam etc. uploading images as people are gradually getting their hands on it and uploading the prompt inputs sent to the lab being processed in about 15 minutes per image with nearly no over-running time. A google lab guy who was also in the chat said that's a improvement over the 45 minutes it takes to procure the hazy AI thoughtprocess images so really it nearly mass-production.

* - A few personal requests from twitter tag-in replies or friends of the dev-team did make them produce questionable anime and fursonas, but all sexual and political content is omitted.

My personal favorite is Otto van Garfield (abstractly posted from reddit):

Spoiler (click to show/hide)

Also quite good mentions

• Thanos versus Harry Potter - (a meme exercise by a dev)

• A realistic love-laugh-life sign

• quote, unquote "Super Mario as Don Corleone"

• Photorealistic pizza with lego topping

Enclosed:

Spoiler (click to show/hide)

[wierd]

I'll wait for this to end up like wombo.ai, with an open prompt.

It should be good for a laugh or two when it is open access.

[None]

So you'll need to find the words to describe the image you want, wait fifteen minutes, find some words to explain the image that was produced, translate those words you need to explain the image, and then present both the conversational partner.

The technology is incredible, but 'as communicative guide' is a solution looking for a problem.

[wierd]

Since the results look fairly polished, it may have uses for places stock art gets used.  15min to generate an image from an obscure text prompt is fine, if you are say, spending an hour writing an opinion article, and just want a humorous image that is royalty free to tag it with.

I could see this being quite useful for such things.

[Scoops Novel]

Look guys, we all have avatars. 5 minutes after this goes mainstream most of our conversation will be them walking and talking (for us).

[MrRoboto75]

Look guys, we all have avatars. 5 minutes after this goes mainstream most of our conversation will be them walking and talking (for us).

"Belongs in the Trash!"

[None]

Hah, no, no it will not, for all the same reasons we don't all communicate over Second Life or facebook's shitty metaverse thing, never mind that again, it's still super computationally expensive and scaling up image generation for not two hundred sample users but millions is beyond feasible.

[Scoops Novel]

Google suggests it would actually be cheaper then the text AI equivalent, GPT-3, which you can already pay for with the cost of a netflix subscription brah.