This AI Can Draw Anything You Describe [Dalle 2 by OpenAI]

What if I told you there is an AI system right now that can take whatever you describe to it, and it will take that text and turn it into a realistic image of exactly what you described. The AI behind this is Dalle 2.
If you can describe it, it can draw it! In this video, we will look at Dalle 2, how it can turn simple text descriptions into realistic images in under 10 seconds, and 10 of its most impressive illustrations. It gets even better, as we will see what Dalle 2 thinks it looks like! Spoiler alert, it appears to be cuddly.

In June 2020 Open AI created GPT 3 a magical AI that could finish your sentences and generate website layouts from a written description. A few months later scientists at open AI thought that If they could complete sentences, why not try to complete images too? And that’s how image GPT was born. Initially, Image GPT was given incomplete images and was asked to fill in the missing pixels.

There are 2 main AI technologies behind Dalle 2, they are called CLIP and DIFFUSION. Clip is the part that matches images to text and uses that match to train computers to understand concepts in images. So when you ask the AI to generate an image of a raccoon playing tennis at Wimbledon in 1990s it knows what a raccoon is and knows what the concept of playing tennis means and it knows what images in the 1990s looked like. And most impressively it knows what’s an aesthetically pleasing image to humans. So it can create a completely new visual version of this idea that has not been created before.
Clip doesn’t have the ability to create high-resolution images by itself. This is where Diffusion comes in. By teaching a computer to corrupt an image by adding Gaussian noise, it can then learn to un-corrupt or enhance an image by removing that noise.

Dalle 2 is the most advanced AI we have ever seen. Ever since the conception of AI, art has been one of the fields that AI has not overtaken or even managed to get a stronghold of. But Dalle 2 has shown impressive results, dominating any other player in the game.

0:00 Intro
1:19 OpenAI
1:33 How DALL.E 2 works
4:07 10 Most impressive photos by DALL.E 2
7:32 DALL.E 2 Editing capabilities
8:18 DALL.E 2 Limitations
9:23 What DALL.E 2 Looks like

.
.
.
.
.
.
.
Copyright Disclaimers • We use images and content by the YouTube Fair Use copyright guidelines • This video could contain certain copyrighted video clips, pictures, or photographs that were not specifically authorized to be used by the copyright holder(s), but which we believe in good faith are protected by federal law and the fair use doctrine for one or more of the reasons noted above.

Credits:
This video would have not been possible without these amazing channels Check out these amazing channels and be sure to support them:

[xarc]

I would like to try weird things that could confuse the AI like: "a man watching a man who is watching another man who is watching a dog who is watching the first man" or "a man made of houses that enters a house made of men"

[EonWhite]

Ask Dall.e 2 to draw an image of:

Planet Earth as a mosaic of smaller images of things that exist on planet Earth.

A tree opening its eyes.

A man with a chainsaw for a head and hands.
(Basically, make it draw Chainsaw Man from the manga of the same name)

A colourful bird made of leaves from a tree.

A cityscape made of many different kinds of plants.

A new species of animal.

A giant robot heroically posing in a city full of humans.

A cute yellow mouse with round red cheeks and a tail in the shape of the symbol for a lightning bolt. Surrounded by electric sparks. It is done in a cartoonish style.
(Basically, make it try to draw a Pikachu)

Draw an ominous and unsettling creature in the dark woods.

Draw a stout and short man with a broad black moustache curving up. He is wearing brown shoes and blue overalls/dungarees and a red shirt with white gloves and a red cap with a red letter “M” within a white circle on the front of the cap in a cartoon style.
(Basically, try to make it draw Mario)

A brown gorilla holding a wooden barrel above its head. On the barrel is a red circle with a bunch of bananas in it.
(Basically, try to make it draw Donkey Kong)

A creature made bones, claws, talons, teeth, black hair, and darkness at the end of a corridor.

A new and complex geometric shape, where each new line is a different colour.


I really wish I could play around with this A.I.. But I have no good reason to, so I won’t bother with the wait list.
I just hope people will make more videos on it instead.

[Bhatt_Hole]

I love his style of narration. It's so,



...fresh!

[asdvdl]

Not sure about it, looks like a fake. Online demo does not work.

[cubeometry]

Try a robot crow flying over a distopian city watching cartoon about utopia.

[dxrk]

I love dall-e videos, but when someone treats us like fucking babies, "nYe nyE cAn tUrN TExT inTo imAgE HaHAHahHahH" I instantly leave

[norbertfeurle]

when i dream, some signs also have scrambled letters like the sign generated from dalle

[Erilis]

I thought Dalle 2 didn't allow realistic human faces. Or is it just celebrities/public figures? Edit: nevermind, found it here: 9:10

[NitwitsWorld]

I do wanna see Dall-E 2 make an ice cream woman in the style of cuphead but she's a sexy anthro pinup back in the 1930s in color

[prophetofthesingularity]

How about
" An android with the face of Isaac Asimov sleeping on a bed in a surreal landscape dreaming of an android with the face of Philip K. Dick that is dreaming of electric sheep."

Shame really that they did not live to see all of this.

[undergroundo]

We should start calling it "Dall-e" instead of "Dall-e 2". The "2" is just the current version. And you can bet there will be plenty more of them.

[xatofoe]

I were thinking of 2 drawings:
An asteroid on the moon watching earth getting destroyed from an asteroid while eating popcorn, vaporwave style.
A flying steampunk fish shooting lasers from its eyes.

[asterlofts]

I would like to create Valve video game crossovers and others such as Halo, with worlds from television series, anime, etc. But I am patient and will wait until the technology is open to everyone and not just by request. Also, by the time it comes out, it will be better implemented.

[xtati]

Do I have to pay if I get 'approved'?

[adityashinde]

" A man frustated by looking at a huge traffic"

[michaelbarry]

A dwarven warrior, an elven ranger and a halfling wizard fighting an undead giant in a dark forest in a style influenced by H.R. Giger and Luis Royo.

[chrisnichols]

What about "A giant mouse destroying Metropolitan city"

[joemacleod-iredale]

lion made of cheese in a bath of lego in the style of studio Ghibli

[wawbutfindable]

try "A cow tyring to breakdance in the motorway" Respond to this comment with a url to the best looking generated image please

[avieda]

ok, try this: "paint the 2008 crisis chart with japanese candlestick in a Salvadore Dali style". thank you!