it turns out that using neural networks to create images is a ton of fun.

recently researchers at google have been training their nets to hallucinate. just last week a team has used a similar technique to replicate artistic style.

and now i’m using those same ideas to mash images together. poor Mona will never be the same again.

many thanks to Kai Sheng Tai for providing his amazing code. sadly it’s written for a CUDA powered system, which i don’t actually have.

so i went over to amazon’s EC2 and rented a beefy system for prototyping. i had to install CUDA drivers and dependencies. luckily amazon makes it trivial for me to share the virtual machine with you: ami-d32a49b6. ENJOY!

(some notes on the system: the neuralart code requires an X display. it takes a couple minutes to generate the 500 iterations for a single image. creating images isn’t a perfect science – i’ve cherry picked the best 10% above.)

---

Filed under: vfx - @ 3 September 2015 6:49 pm