By analyzing various artists’ distinctive treatment of facial geometry, researchers in Israel devised a way to render images with both their painterly styles (brush strokes, texture, palette, etc.) and shape. Here’s a great six-minute overview:

Things the internet loves:
Nicolas Cage
Cats
Mashups

Let’s do this:

Elsewhere, I told my son that I finally agree with his strong view that the live-action Lion King (which I haven’t seen) does look pretty effed up. 🙃

Hard to believe that it’s been almost seven years since my team shipped Halloweenify face painting at Google, and hard to believe how far things have come since then. For this Halloween you can use GANs to apply & animate all kinds of fun style transfers, like this:

I dunno, but it’s got me feeling kinda Zucked up…

FaceMix offers a rather cool way to create a face by mixing together up to four individually editable images, which you can upload or select from a set of presets. The 30-second tour:

Here’s a more detailed look into how it works:

Generative artist Glenn Marshall has used CLIP + VQGAN to send Radiohead down a rather Lovecraftian rabbit hole:

“A nuclear-powered pencil”: that’s how someone recently described ArtBreeder, and the phrase comes to mind for NVIDIA Canvas, a new prototype app you can download (provided you have Windows & beefy GPU) and use to draw in some trippy new ways:

Paint simple shapes and lines with a palette of real world materials, like grass or clouds. Then, in real-time, our revolutionary AI model fills the screen with show-stopping results.

Don’t like what you see? Swap a material, changing snow to grass, and watch as the entire image changes from a winter wonderland to a tropical paradise. The creative possibilities are endless.

[Via]

A few weeks ago I mentioned Toonify, an online app that can render your picture in a variety of cartoon styles. Researchers are busily cranking away to improve upon it, and the new AgileGAN promises better results & the ability to train models via just a few inputs:

Our approach provides greater agility in creating high quality and high resolution (1024×1024) portrait stylization models, requiring only a limited number of style exemplars (∼100) and short training time (∼1 hour).

Generative artist Nathan Shipley has been doing some amazing work with GANs, and he recently collaborated with BMW to use projection mapping to turn a new car into a dynamic work of art:

I’ve long admired the Art Cars series, with a particular soft spot for Jenny Holzer’s masterfully disconcerting PROTECT ME FROM WHAT I WANT:

Here’s a great overview of the project’s decades of heritage, including a dive into how Andy Warhol adorned what may be the most valuable car in the world—painting on it at lightning speed:

Years ago my friend Matthew Richmond (Chopping Block founder, now at Adobe) would speak admiringly of “math-rock kids” who could tinker with code to expand the bounds of the creative world. That phrase came to mind seeing this lovely little exploration from Derrick Schultz:

Here it is in high res:

I’ve obviously been talking a ton about the crazy-powerful, sometimes eerie StyleGAN2 technology. Here’s a case of generative artist Mario Klingemann wiring visuals to characteristics of music:

Watch it at 1/4 speed if you really want to freak yourself out.

Beats-to-visuals gives me an excuse to dig up & reshare Michel Gondry’s brilliant old Chemical Brothers video that associated elements like bridges, posts, and train cars with the various instruments at play:

Back to Mario: he’s also been making weirdly bleak image descriptions using CLIP (the same model we’ve explored using to generate faces via text). I congratulated him on making a robot sound like Werner Herzog. 🙃

The Avalanches used ArtBreeder (see previous) to make this trippy, ever-morphing music video.

This provides me a periodic reminder that I’ve never seen What Dreams May Come & should someday fix that.

Artbreeder is a trippy project that lets you “simply keep selecting the most interesting image to discover totally new images. Infinitely new random ‘children’ are made from each image. Artbreeder turns the simple act of exploration into creativity.” Check out interactive remixing:

Artbreeder is a nuclear powered pencil.

have a spin remixing some of mine here:https://t.co/pjeX7PNcgC#artbreeder #ai #ganbreeder #conceptart #comics pic.twitter.com/zxGLculJtA

— Bay Raitt (@bayraitt) September 17, 2019

Here’s an overview of how it works:

Generative Adversarial Networks are the main technology enabling Artbreeder. Artbreeder uses BigGAN and StyleGAN models. There is a minimal open source version available that uses BigGAN.

I find this emerging space so fascinating. Check out how Toonify.photos (which you can use for free, or at high quality for a very modest fee) can turn one’s image into a cartoon character. It leverages training data based on iconic illustration styles:

I also chuckled at this illustration from the video above, as it endeavors to how two networks (the “adversaries” in “Generative Adversarial Network”) attempt, respectively, to fool the other with output & to avoid being fooled. Check out more details in the accompanying article.

On Monday I mentioned my new team’s mind-blowing work to enable image synthesis through typing, and I noted that it builds on NVIDIA’s StyleGAN research. If you’re interested in the latter, check out this two-minute demo of how it enables amazing interactive generation of stylized imagery:

This new project called StyleGAN2, developed by NVIDIA Research, and presented at CVPR 2020, uses transfer learning to produce seemingly infinite numbers of portraits in an infinite variety of painting styles. The work builds on the team’s previously published StyleGAN project. Learn more here.