Sunday, May 26, 2019

Fake Fashion Models

Computers can create photo-real images of fashion models that never lived in the flesh.



In the short morphing video, hemlines rise and fall, stances shift, and genders blend into each other. (Link to video on YouTube)



The developers in Japan use generative adversarial networks to create the images. They have also set up their AI system to generate fake pop stars.


They say they plan to improve the technology to the point they can offer it to the apparel, cosmetics, and advertising industries.


They're also working with human co-creators to produce new Manga-style characters, saying the technology will help them "to significantly reduce the cost of producing content."
----
Data Grid company
Via BoingBoing



10 comments:

Eugene Arenhaus said...

*looks at the middle image* Say hi to eyes with a cast, punched-through ears with hairs, and wavy cheek lines...

These guys, along with most AAA computer game makers and a lot of artists I know, should be forced to write 10,000 times on the blackboard: "Photorealistic rendering does not improve shoddy construction."

Luca said...

I believe we are really close to a point in which is impossible to tell if an actor is real or a CGI.
But i see this news as a bit more disturbing, for everything its consequences:

https://www.sciencealert.com/samsung-s-ai-can-now-generate-talking-heads-from-a-single-image

What do you think about it?

Luca said...

PS: i realize this technology is still very rough and results not very precise, but...

James Gurney said...

Luca and Eugene, yes, the image isn't perfect yet, but to be fair, I grabbed a frame from a morphing video. I think it's obvious that computer-created images will soon be indistinguishable from photographs or human-created paintings.

What I'm curious to see is what AI can come up with that doesn't try to simulate art or photography, but rather that pursues a more open-ended mode of imagery not tied to anything but the response of humans....imagery shaped directly by biofeedback of human viewers....externalized hallucination custom tailored to the individual. Art will then become not a thing we look at together, but something we experience separately within our own minds, assisted by machines.
Edit or delete this

Terry said...

Do. Not. Want.

Celia said...

What Terry said!

Unknown said...

"They're also working with human co-creators to produce new Manga-style characters, saying the technology will help them "to significantly reduce the cost of producing content."

This reads to me as a euphemism for, "We won't have to hire models, photographers, or run photo shoots anymore.

PeggyB said...

Many years ago, Time Magazine did a cover with a woman's very attractive face on it. This was in the early days of morphing, but they had taken photographs of a woman from every race in the world, and morphed them all together. The hair was pulled back from the face pretty severely, essentially removing that element.
It was a lesson in morphing, but more than that they said that ultimately, should we survive all damage and foolishness, this was what the human race would look like. She was not stunning, she was lovely. "Perfect," even.

Terrace said...

The thing about this technology -- like all technology -- is that eventually, audiences develop an understanding and a resistance to it. When I was younger, CGI in films used to astonish me and it seemed very real. Nowadays, I'm much more inured to the effects of CGI and much more skilled at picking out the "uncanny valley" of it, no matter how subtle. There is a certain "CGI look" that these films give off that creates boredom in a knowing audience. As for these morphs, I can see they have certain subtle properties that will become more and more familiar the more they are used in the visual industries. In the future, canny audiences will be able to instantly look at these computer-generated models and understand that they are computer-generated.

And as for manga being produced by computer... well, no great loss there, it will free up young artists to try their hands at more comprehensive art study :-)

Eugene Arenhaus said...

"What I'm curious to see is what AI can come up with that doesn't try to simulate art or photography, but rather that pursues a more open-ended mode of imagery not tied to anything but the response of humans"

You'll just get hallucinatory imagery, I expect. As confusing and repetitive as the real thing. Though it would be an interesting experiment.

As such, this kind of "AI" cannot generate novel things, just regurgitate variations on its training set, and is unlikely to move further than that without serious breakthroughs in the algorithm.