OpenAI's New AI Model Draws Images From Text

The machine learning company OpenAI is developing models that improve computer vision and can produce original images from a text prompt. From a report: The new models are the latest steps in ongoing efforts to create machine learning systems that exhibit elements of general intelligence, while performing tasks that are actually useful in the real world -- without breaking the bank on computing power. OpenAI this week is announcing two new systems that attempt to do for images what its landmark GPT-3 model did last year for text generation. DALL-E is a neural network that can "take any text and make an image out of it," says Ilya Sutskever, OpenAI co-founder and chief scientist. That includes concepts it would never have encountered in training, like the drawing of an anthropomorphic daikon radish walking a dog. DALL-E operates somewhat similarly to GPT-3, the huge transformer model that can generate original passages of text based on a short prompt. CLIP, the other new neural network, "can take any set of visual categories and instantly create very strong and reliable visually classifiable text descriptions," says Sutskever, improving on existing computer vision techniques with less training and expensive computational power.

Hmmm

[raymorris]

"take any text and make an image out of it"

How can I try this out?
"Halle Berry, Camilla Cabello and me in bed"

Re:

[Tablizer]

Division by zero. Length processing error while drawing Ray's romantic-category anatomy.

Re:

[JaredOfEuropa]

I bet something like that was among the first things they tried out. Like that South Park episode with the Internet Porn Simulator: "Two Japanese girls puking in each other's mouths: click"

Re:

[greytree]

I tried it.
Didn't work.
The AI needs to know where Halle Berry and Camilla Cabello were while you were in bed.

AI creap-outs will increase

[Tablizer]

The first can of Skynet Soup looks absolutely disgusting.

Re:

[Iamthecheese]

"delicious-looking" wasn't specified. gigo

Re:

[Tablizer]

Maybe it was, but trained on 4chan content.

Re:

[greytree]

Yes, but it can form a liquid metal spike and open itself.

Then destroy humankind.

Full circle

[Scarred Intellect]

DALL-E is a neural network that can "take any text and make an image out of it,...CLIP, the other new neural network, "can take any set of visual categories and instantly create very strong and reliable visually classifiable text descriptions,"

I only have two questions:

Can DALL-E rebuild CLIP's input from CLIP's output?

Can CLIP rebuild DALL-E's input from DALL-E's output?

Artist

[kyoko21]

Vincent Van Deep GOgh.

100% pun intended.

How many words?

[An0nYm0u5c0wArD]

Does it take 1000 words to generate an image?

Re:

[An0nYm0u5c0wArD]

That whooshing sound you may hear is the joke going over your head.

For the Police?

[nospam007]

So they can fire all the sketch-artists?

Been using this for 27 years.

[Foundling]

It's called "POV-Ray".

Re: Been using this for 27 years.

[Tablizer]

But it always says "Syntax error near 'sucking my'."

Never again

[systemd-anonymousd]

Oh yes, "Open"AI, who is so open that once GPT-2 got popular they created an API, paywalled it, called it GPT-3, and bound you to restrictive terms of service that will cut your application off if you do anything too political incorrect. How "Open" of them.

Porn anyone?

[BAReFO0t]

Come on, imagine extending it to video, and the giving it a Siri interface.

Cease and Desist

[AlejandroTejadaC]

"an illustration of a pikachu in a cape wielding a blue lightsaber" Nintendo, DC Comics and LucasFilms lawyers want to talk with you...

Wonderful news!

[UBfusion]

I've been waiting years for this: Alexa, draw me a nice cock and balls!

Re:

[Tablizer]

It drew two politicians.

Can it draw:

[greytree]

Goat ... something ?

Re:

[Tablizer]

On behalf of 90% of slashdotters, Nnnoooo!