"generative adversarial networks"

Astonishingly weird video of AI-generated facial expressions mapped to music

AI artist Mario Klingemann used Generative Adversarial Networks (GANs), one of the primary techniques to create deepfake videos, to make this incredible, unsettling, and wonderful video that facial expressions to music. (Song: "Triggernometry" by Kraftamt, 2014). Check out another deepweirdfake from this series below.

(Thanks, Jeff Cross!) Read the rest

GANfield: ai-generated Garfield

GANField is the result (nightmarish, grotesque, occasionally comical) of a generative adversarial network trying to find its way through a corpus of Garfield strips.

Watch as an AI struggles to seam together hundreds of Garfield comics. Created by Daniel Hanley. https://twitter.com/calamardh Music is '4 All Hail the Fishmen' from LISA: The Painful

You have no idea how ... unlonely you are, Garfield!

See also: Will Burke's creepy Garfield drawings. Read the rest

Hopefully This Cat Does Not Exist

You've seen This Cat Does Not Exist, now behold Hopefully This Cat Does Not Exist, whereby the failings of generative adversarial networks become their horrifying strengths.

I'm a big fan of the ones where GANs try to write meme captions and ends up with this ... abyssal syllabary. AI text generation naturally focuses on creating readable phrases, but training it on the appearance of text is where the cold magic springs.

Read the rest

Celebreedy uses AI to combine faces of stars

Celebreedy breeds new celebrities from other celebrities, using a generative adversarial network. It's the work of Eric Drass, who has a website.

Eric holds a degree in Philosophy and Psychology (Oxford) and an unfinished PhD in Cognitive Psycholinguistics (also Oxford). He is co-author on a number of patents dealing with PRISM-type surveillance technologies (long before PRISM became public), and a number of academic papers relating to neural network models of language acquisition and heritability.

He also used to be a singer in an experimental hardcore band, an unsuccessful male model, and once took at dotcom 1.0 company from a bedroom project to 14 countries and back, spending $50m on the way. Twenty years ago he was a TV star in America, but he doesn’t like to talk about it.

Read the rest

AI generates old-fashioned zoological illustrations of beetles

These beetles do not exist: Confusing Coleopterists is an AI trained on illustrations from zoological textbooks. The extreme formality of this art genre, and its placement within the public domain, makes it uniquely apt to the medium of generative adversarial networks: "Results were interesting and mesmerising."

I set up a machine at PaperSpace with 1 GPU (According to NVIDIA’s repository, running StyleGan on 256px images takes over 14 days with 1 Tesla GPU) ?

I trained it with 128px images and ran it for 3 days, costing €125.

Results were nice! but tiny. ... I loaded the beetle dataset and trained it at full 1024px, [on top of the FlickrHD model] and after 3000 steps the results were very nice.

No-one below Ph.D. level should ever trust an illustration of a beetle again! Read the rest

This company wants to use AI to help you pretend to increase diversity

Generated Photos is the latest stupid startup that sounds like a joke from "Silicon Valley" that someone took too far. From their announcement on Medium:

Generated Photos is the free resource of 100k faces for you to use however you wish. But these aren’t just common faces. They were produced completely by artificial intelligence — none of these people are real! Generated photos are created from scratch by AI systems.

In other words, they're Deepfakes for other peoples' ad campaigns.

I've spent enough time around higher ed administration that I've seen firsthand how universities will recruit a perfect United-Colors-Of-Benetton rainbow of students for admissions ads. But this takes that to a whole new level. Why even bother trying to build relationships with non-white-dudes, when you can just generate some friendly colorful faces for promotional use and call it a day?

The company's website brags of "democratizing creative photography and video," which is some impressively nauseating PR speak. In their defense, "We aim to make creative works both more accessible and higher quality through generative processes" sounds a lot better than "Auto-diversify the avatars for your army of Twitter sockpuppets!"

But my favorite part is how openly they acknowledge the poor quality of their images. "A part of the process is training and refining the generative models," the company explains in a Medium post. "The iterations move fast although not everything is perfect yet. So you will also have some fun with the pack of AI-generated photos. When you see a face that is a bit ‘off’, just give it some slack." Read the rest

Watch an AI-generated human face become monstrous as its neural network decays

"What I saw before the darkness" shows a human face created by a generative adversarial network, next to the neural memory map representing it. Neurons are removed, and as they go the face loses detail, becomes vague, and finally decays to something that may speak ill of the machine and the mankind that made it.

Memories disappear unwitnessed. Images gradually fade away Until one day there is nothing left But a vague feeling of loss…

Here's the project page: AI Told Me.

Vice's Samantha Cole:

"The inspiration behind the project is rooted in contemplation of human perception," the creator said. "Everything we see is the brain’s interpretation of the surrounding world. A person has no access to the outside reality except for this constructed image."

She compared this to how Claude Monet's paintings shifted to blurred, muddled greens and yellows as he aged: Our eyes and brains and the networks that connect them undergo changes and deterioration that we might barely notice as it’s happening.

Read the rest

Company uses AI to generate whole-body images of people who don't exist

Datagrid says, "We have succeeded in generating high-resolution (1024×1024) images of whole-body who don't exist using Generative Adversarial Networks (GANs). We use these images as virtual models for advertising and fashion." Read the rest

Creative Adversarial Networks: GANs that make art

Generative Adversarial Networks use a pair of machine-learning models to create things that seem very realistic: one of the models, the "generator," uses its training data to make new things; and the other, the "discerner," checks the generator's output to see if it conforms to the model. Read the rest

This website is a Hot-or-Not for fake people

Mike Solomon, creative director at Hearst Digital Media, created Judge Fake People. Here's what he wrote about it on his website, The Cleverest:

Like many internet addicts, I was blown away by NVIDEO’s demo using style-based Generative Adversarial Networks to generate faces. They seem to have crossed a threshold for generating artificial images that can genuinely fool our brains.

Flash forward to last week when I saw Philip Wang’s amazing single-serving website thispersondoesosxist.com.

It was fascinating to see a photo-realistic face of someone that doesn’t exist. Philip’s writeup does a great job explaining his motivations and this implications behind this groundbreaking technology.

But I just wanted to turn it into hot or not.

So I wrote a script to download an image from thispersondoesnotexist.com every 5 seconds and built up a collection of around two thousand fake people. Then I made a voting system with php/MySQL and some filters to show the highest and lowest rated faces. And I enabled comments just for fun.

Read the rest

How to recognize AI-generated faces

First, take this quiz to see how good you are at distinguishing between real faces and fake ones made with generative adversarial networks (GANs). Then, read this article that teaches you how to spot the fakes. In a few years AI will be able to generate images that don't have recognizable tells. Read the rest

Breed weird critters with machine learning and Ganbreeder

Ganbreeder uses a machine learning technique called Generative Adversarial Networks (GANs) to generate images that seem like photos, at least a first glance. Read the rest

Generative adversarial network produces a "universal fingerprint" that will unlock many smartphones

Researchers at NYU and U Michigan have published a paper explaining how they used a pair of machine-learning systems to develop a "universal fingerprint" that can fool the lowest-security fingerprint sensors 76% of the time (it is less effective against higher-security sensors). Read the rest

DeOldify: a free/open photo-retoucher based on machine learning

Jason Antic's DeOldify is a Self-Attention Generative Adversarial Network-based machine learning system that colorizes and restores old images. It's only in the early stages but it's already producing really impressive results, and the pipeline includes a "defade" model that is "just training the same model to reconstruct images that augmented with ridiculous contrast/brightness adjustments, as a simulation of fading photos and photos taken with old/bad equipment." Read the rest

This AI-generated portrait just sold at auction for $432K, but not without controversy

This blurry portrait of a man may not look like much but it just sold at auction for $432,500, nearly 45 times its high estimate. What makes it so special? The Portrait of Edmond Belamy is the work of Artificial Intelligence and it's the first of its kind to sell at a major auction house.

Christie's:

This portrait, however, is not the product of a human mind. It was created by an artificial intelligence, an algorithm defined by that algebraic formula with its many parentheses. And when it went under the hammer in the Prints & Multiples sale at Christie’s on 23-25 October, Portrait of Edmond Belamy sold for an incredible $432,500, signalling the arrival of AI art on the world auction stage.

The painting, if that is the right term, is one of a group of portraits of the fictional Belamy family created by Obvious, a Paris-based collective consisting of Hugo Caselles-Dupré, Pierre Fautrel and Gauthier Vernier. They are engaged in exploring the interface between art and artificial intelligence, and their method goes by the acronym GAN, which stands for ‘generative adversarial network’.

‘The algorithm is composed of two parts,’ says Caselles-Dupré. ‘On one side is the Generator, on the other the Discriminator. We fed the system with a data set of 15,000 portraits painted between the 14th century to the 20th. The Generator makes a new image based on the set, then the Discriminator tries to spot the difference between a human-made image and one created by the Generator.

Read the rest

Enhance enhance: Using machine learning to recover lost detail from upscaled photos

A team of researchers from Twitter have published a paper detailing a machine learning technique that uses a generative adversarial network to make shrewd guesses about how to up-res small images by up to 400%, into crisp, large images, with eye-popping results. Read the rest

Hypnotic video of imaginary celebrities generated by a neural net

A generative adversarial network (GAN) combines two neural networks engaged in a zero-sum competition. The result is a form of unsupervised machine learning that can produce imaginary celebrities like the ones shown in this one-hour video. Read the rest

:)