[Much of the coverage of the latest advancement in the ability of AI to generate presence-evoking images – this time of fashion models who don’t exist – points to the potential dangers. The story below is from Futurism (where you can watch the 0:44 minute demo video), and here’s a quote from BGR’s coverage: “This technology is impressive, but at the same time it also seems to herald something that humanity still has not fully come to grips with yet. It’s only a matter of time before AI masters the creation of fake human speech and other things that, along with the images, can be integrated into a full package and really give us a hard time with separating fact from fiction. Maybe someone will develop an AI one day that masters the art of fighting disinfo and ferreting out fake news. We can only hope.” –Matthew]
[Image: Source: CTV]
Amazing AI Generates Entire Bodies of People Who Don’t Exist
The algorithm whips up photorealistic models and outfits from scratch.
Dan Robitzski
April 30th 2019
Embodied AI
A new deep learning algorithm can generate high-resolution, photorealistic images of people — faces, hair, outfits, and all — from scratch.
The AI-generated models are the most realistic we’ve encountered, and the tech will soon be licensed out to clothing companies and advertising agencies interested in whipping up photogenic models without paying for lights or a catering budget. At the same time, similar algorithms could be misused to undermine public trust in digital media.
Catalog From Hell
The algorithm was developed by DataGrid, a tech company housed on the campus of Japan’s Kyoto University, according to a press release.
In a video showing off the tech, the AI morphs and poses model after model as their outfits transform, bomber jackets turning into winter coats and dresses melting into graphic tees.
Specifically, the new algorithm is a Generative Adversarial Network (GAN). That’s the kind of AI typically used to churn out new imitations of something that exists in the real world, whether they be video game levels or images that look like hand-drawn caricatures.
Photorealistic Media
Past attempts to create photorealistic portraits with GANs focused just on generating faces. These faces had flaws like asymmetrical ears or jewelry, bizarre teeth, and glitchy blotches of color that bled out from the background.
DataGrid’s system does away with all of that extraneous info that can confuse algorithms, instead posing the AI models in front of a nondescript white background and shining realistic-looking light down on them.
Each time scientists build a new algorithm that can generate realistic images or deepfakes that are indistinguishable from real photos, it seems like a new warning that AI-generated media could be readily misused to create manipulative propaganda. Here’s hoping that this algorithm stays confined within the realm of fashion catalogs.