It would therefore be a great shame if someone offered infinite free versions of the collectibles — which is exactly what Yannic Kilcher has done. The machine learning expert has built an app that generates as many Bored Apes as your heart desires. The system is powered by generative adversarial networks (GANs), which create new content by pitting two neural networks against each other: a generator that produces new images and a discriminator that guesses which ones are fake. Over time, the generator starts creating new pictures that can pass as authentic. Kilcher first scraped the Bored Ape Yacht Club collection of 10,000 images. He then trained Nvidia’s StyleGAN2 neural network on the dataset. The finished system produces Bored Apes that are entirely unique — and fungible. Kilcher admits, however, that the images aren’t always perfect. “The quality of my system is not very good, there are often visible artifacts,” he told TNW. He adds that Cyril Zakka and Nathan Cooper Jones have built similar tools that produce better outputs. Yet Kilcher has also created something truly original: an application that generates apes based on images that you upload. The app harnesses OpenAI’s CLIP — the same model that powers the DALL-E 2 image generator. You can try it out for yourself, but you’ll need a local GPU to run the system. The standard Bored Ape generator, meanwhile, can be used by any simian-lover who visits Hugging Face or Kilcher’s website. Our thoughts and prayers go out to Jimmy Fallon and Paris Hilton at this difficult time.
