(Model & site by Gwern Branwen.)
Do Electric Neural Nets Dream Of Anime Shows?
[Refresh for a random deep learning StyleGAN 2-generated anime face & GPT-3-generated anime plot; reloads every 15s. For many waifus simultaneously in a randomized grid, see "These Waifus Do Not Exist". This website's images are available for download. For interactive waifu generation, you can use Artbreeder which provides the StyleGAN 1 portrait model generation and editing, or use Sizigi Studio's similar "Waifu Generator". You may also enjoy "This Fursona Does Not Exist".
Technical note: these faces were generated with 𝜓=0.8 for images #0–50,000 (medium quality/medium diversity), 𝜓=0.6 for #50,001–75,000 (high quality, low diversity), and 𝜓=1.1 for #75,001–100,000 (low quality, high diversity) for a mix of good & interesting faces; for details about the 𝜓 hyperparameter, see the StyleGAN 1 paper. Pre-v3 images are still available at
The "plot summaries" were generated by the GPT-3 API NN (175b). For more details, see the TWDNE site implementation writeup and my StyleGAN training tutorial. (Note: TWDNE should not be confused with "This Word Does Not Exist".)]