Hacker News Reader

Hacker News

Comment by plibither8 | Hacker News Reader

Parent Full thread

plibither8

A few more...

* Person: https://thispersondoesnotexist.com/

* Waifu: https://www.thiswaifudoesnotexist.net/

* SO Question: https://stackroboflow.com/

* Startup: https://thisstartupdoesnotexist.com/

* Resume: https://thisresumedoesnotexist.com/

View on HN

There used to exist https://thisairbnbdoesnotexist.com/ [1] Oh the irony...

[1] On Archive.org - https://web.archive.org/web/20190220071550/https://thisairbn...

The Waifu one is so realistic that it's uncanny. It's almost like we're just a few years away from whole seasons of deep-dreamt anime and whole tankoubons of procedurally generated manga.

Blurb generated by GPT-2 for TWDNE #283: https://www.gwern.net/images/gan/thiswaifudoesnotexist-283.p...

“It is so sad to say that this manga has never been seen by any anime fans in the real world and this is an issue that must be addressed. Please make anime movies about me. Please make anime about me. Please make anime about your beautiful cat. Please make anime movies about me. Please make anime about your cute cat. I wish you the best of luck in your life. Please make anime about me. Please make anime about my cute cute kitten.”

Maybe it's just a prank, but Gwern's description makes it sound like an accidental creation of the text generator. https://www.gwern.net/TWDNE

Nope, it's real: that snippet is just that hilariously on point. To be fair, that's one of the best snippets out of many thousands. Few are anywhere that amusing.

The images & snippets are 100% uncurated & unedited by a human, other than as described in my page (primarily: I feed in a long prompt to give it keywords to work with, and I postprocess them to drop at '<endoftext>' tokens to keep them on a single topic rather than, as the default generation process does, force starting of a new topic to fill out the character count).

This text just sounds like the stuff that I used to get in Akismet when I had a Wordpress blog.

I mean, afaik the manga/anime style specifically emerged from manga being drawn as fast as possible, as a co-evolution with the thick tomes of magazines. So I'd expect drawing a face to not be a bottleneck there.

It's going in the wrong direction. Sure, it can conjure up a random picture of a face. But for purposes or manga/anime building, it's near useless, as it can't generate that face in multiple perspectives, or with different kinds of facial expressions. This Waifu truly does not exist - neither in prior publications, nor in the algorithm itself.

That's incorrect. You can generate the same face with multiple facial expressions by either doing style transfer onto a single face from other faces using the style noise (as demonstrated in the StyleGAN paper & video, and you can see further examples with my anime StyleGAN faces), or you can directly control a face by modifying its latent vector as desired, once you've reverse-engineered the latent vector (as can be done by generating random faces+latents, & feeding the faces into a tagger like DeepDanbooru which knows about facial expressions to do classification/regression). See https://www.gwern.net/Faces#reversing-stylegan-to-control-mo...

For a recent cool example, http://everyoneishappy.com/portfolio/waifu-synthesis-real-ti... controls the mouth for sing-along syncing.

Wow. Thanks for that; I'm very happy to have been wrong about this issue, and I've just learned a lot from reading your Faces page.

One issue I notice in your video of interactive face editing is what I'd call "crosstalk" (there's probably a more formal term for this) - you alter a specific slider, and beyond the targeted modification, the entire face, shirt and background changes slightly. Is there any work on eliminating such crosstalk, so one could e.g. stitch together a believable video from such dynamically modified face?

My theory is that the latent space isn't perfectly 'disentangled' (there's not really anything in StyleGAN akin to InfoGAN which would force it to be exactly linear/disentangled, it just nicely works out roughly disentangled), so when you train a linear model on the tags+latent and manipulate it linearly, you get slight changes in correlated features. Thus, the slight 'glitchiness'. Not something you'd notice in image samples, but more apparent when you generate interpolations of the 'same' face.

If you mess with the style noise only and leave the latent vector alone, the glitchiness should be less, but you also lose a lot of possible control. Training a more powerful nonlinear classifier like a random forests should be able to compensate for the remaining nonlinearity in the latents, I think, but then it's harder to do the actual modification of the latent vector, of course, since you're no longer simply tweaking one variable but many simultaneously to try to hold the other features constant. One could try to brute force it by doing blackbox optimization of the nonlinear classifier... but I haven't seen anyone try that.

Not directly but there are other networks that can do those parts.

I'd like to learn more. Anything you could link to?

https://youtu.be/p1b5aiTrGzY

One shot training examples at 4:23

Reply on news.ycombinator.com