Janelle Shane @JanelleCShane Research Scientist in optics. Plays with neural networks. Avid reader, writer, and player of Irish flute. she/her. wandering.shop/@janellecshane Aug. 16, 2018 1 min read

It turns out "draw me a picture of whatever I ask for" is a really hard task for an algorithm. This one will try its best though:  http://t2i.cvalenzuelab.com/ 
Demo by @c_valenzuelab of this model:  https://github.com/taoxugit/AttnGAN/ 

Anytime this GAN tries to draw humans, it's horrifying because it can never figure out how many orifices there are.

At least when asked to draw something other than humans, the GAN was less terrifying. I rather like this one.

This AI drawing was nice too! If I squint I can maybe see the bears carrying their bear-propulsion apparatus.

Just as I suspected.

But when I ask for *a* sheep then it delivers... and makes the entire planet sheep-textured just to be safe.

It definitely knows giraffe texture. A bit fuzzy on their actual shape.

I really like what it does with clock towers. Although I think that second one might actually be a tower of clocks.

If you ask the GAN to combine things that it already can draw pretty well, it actually does a pretty convincing job.

You can follow @JanelleCShane.


Tip: mention @threader_app on a Twitter thread with the keyword “compile” to get a link to it.

Enjoy Threader? Sign up.

Threader is an independent project created by only two developers. The site gets 500,000+ visits a month and our iOS Twitter client was featured as an App of the Day by Apple. Running this space is expensive and time consuming. If you find Threader useful, please consider supporting us to make it a sustainable project.