How Dall-E 2 and Other AI Art Generators Create Images From Text | WSJ

New text-to-image generators powered by artificial intelligence, including OpenAI Dall-E 2 and Stability AI DreamStudio, let you type in almost any phrase and get an image.

WSJ’s Joanna Stern explains the tech and pits it against a real photographer and a real man in a robot suit.

Tech Things With Joanna Stern
Everything is now a tech thing. In creative and humorous videos, WSJ senior personal tech columnist Joanna Stern explains and reviews the products, services and trends that are changing our world.

More from the Wall Street Journal:

#AiArt #DallE2 #WSJ

[JJs_playground]

I really like the way this reporter does her videos. Keep it up. WSJ giver her a raise.

[hybridroid]

These AI evolve extremely fast. For instance, StableDiffusion is now opensource which means anyone can contribute in order to upgrade it... soon the internet will be flooded with 'fake' images 100% AI generated but the good thing is anyone has now the opportunities to express their ideas beyond our own imagination.

[jdzajdza]

I really enjoy your report and videos. You never disappoint

[brooklynrobotworks]

Well Done! This shoot was A Blast. -Peter

[TTTrouble]

This was a very surface level video and doesn’t do the full capabilities of either Dall-E or stable diffusion Justice. It may take a little bit of adjustment of settings and prompt, but photorealistic images are possible within limits and much more.
Also the ethics of using people’s art styles is also a fascinating topic that deserves to be part of the conversation(Greg Rutkowski)

[SamirMishra]

Jonna Strrn Does it again 🎉 great reporting

[nickgreen]

The newest version of midjourney makes all these photos look like an early 2000s jpeg. It's insane how fast this is evolving.

[wsippel]

a) not censoring or filtering anything is what makes Stable Diffusion the best in the industry, in part because of b): people should never blindly believe anything they see on the internet in the first place. Right now, I can make more convincing fakes in Photoshop, and nobody asks for Photoshop to be regulated. Image synthesis makes it even easier, sure, but maybe that's the wakeup call people need. Stop blaming the tools, blame the bad actors misusing them, and blame yourself for blindly believing nonsense because of your personal bias.

[Magicwaterz]

So, it still uses existing images as basis to generate the art. But each of those images are labeled so that the AI can reference what that word is, and analyzes what do these images in a certain word have in common. My question now is, where do they get these images for the AI to reference. Was it sourced fairly?

[bhavik_sarkhedi]

Fast forward 6 months and the world has completely changed. the way mid journey and stability ai's DreamStudio is moving, the time has come when it will be pretty difficult to figure out which is what - ai or real

[therealjoshschannel]

0:45 0:52 It looks like a Tesla Bot Silver Version
Thumbnail, too

[MAG]

This is quality content. A for effort.

[nicholaslandolina]

She's trying to get proof the man was on her pool deck lol

[maltimoto]

How does the AI generator get the 3D model for the scene? It is fed with billions of 2D photos, are they converted to a 3D model or how does it work?

[akshatjainbafna]

Their are models that inspect the image and tells you weither it is CGI or real image

[EvolvedSungod]

I think a lot of fear artists have comes from the failure of sites like this to explain what algorithms are doing in the initial stages, how they are trained, and what is actually happening in each step to alter the image file. So far I haven't seen a single site explain that stuff as more than "ai is trained on many images" so then artist take that as "its stealing pieces of my art" when it isn't doing that at all. I'm honestly tired of explaining this stuff to angry idiots on social media.

Please explain that part better. Even lawyers and media companies don't seem to understand this aspect of these image generators, further feeding the fear of artists.

[lll-xonk]

he's not a darlec
but a spacemen
...

[auro]

how? by making something that you didn't ask for because the artificial intelligence won't work unless you pay like wsj

[auro]

or wsj can make more money than both by selling that robot costume

[dpainter]

I dislike the use of the terms "learn" and "know" in speaking of AI. AI is not human, it does not "learn" that a balloon is round, it simply "categorises" a balloons as an object which is round.