Despite that the machine does not have a GPU it is able to generate 128x128 pixel #BigGAN images at the blazing speed of 30/minute. 😆

Another descent which was kind of lucky with the start vector. And yeah - as we know faces are not BigGAN's forte. #clip #biggan #FindingMona

"Still Life with Sour Grapes and Pitchfork"

#CLIP #BigGAN

Dogs welcome, but stay away from bridges. #BigGAN #U2Net

#BigGAN #U2Net #Done

It's now doodling from a mixture of random video frames and #BigGAN. I also replaced the long-winded eraser routine with a one-step fill, so it's a bit less challenging for the instant gratification generation.

I made a new thing called "Sketchflix".

Based on @mrdoob@twitter.com's multiuser-sketchpad I let an #aiart bot do the drawing using #BigGAN images as its inspiration. And you can watch live and scribble along (until it breaks).

Here you go: https://sketchflix.glitch.me/

A few more.
[#Collagenerator vs. #BigGAN]

#Collagenerator vs. #BigGAN

Making ends meet. (No comparison to what @mtyka@twitter.com is currently cooking on the #GANorama front.) #BigGAN

Latent Core Drilling. #BigGAN

Inverse #BigGAN search for a previously generated image across the entire 1128-dimensional latent space starting from a random vector.

Not sure it it matters but all the tiles are unaltered #BigGAN results found via 1000 steps of gradient descent towards a target image, and no color adjustments or other post-processing was applied.

Almost done. "On the Shoulders of Giants" #InverseCollage #BigGAN

Into the latent #BigGAN jungle again #FindingMona

I also think that found a pretty way how to interpolate smoothly through fixed landmarks in 1128 dimensional space without kinks in the path.

Aaah, I finally realized why they appear different: this model is using the new and improved #BigGANDeep and not the old #BigGAN.

Interesting, @Thom_Wolf@twitter.com's Pytorch port of #BigGAN seems to make slightly better faces than the original version. But I am probably just imagining that, since the weights are identical.

Can't say that I am a big fan of this type of attention economy art, when "controlling by mind" translates into "appropriating other people's #BigGAN discoveries without giving credit". And accusing people of harassment when called out is not really the appropriate response.

A friendly reminder that Ganbreeder is just one drilling platform into #BigGAN's treasures. So whatever you pull out there might already be claimed by some independent prospectors from the other end of the latent space.