CoSupport AI
What is This Fursona Does Not Exist?
This Fursona Does Not Exist is a website showcasing an endless grid of furry portraits generated by Artificial Intelligence. It utilizes nVidia's StyleGAN2 architecture, a sophisticated AI model trained specifically for this purpose. The training process involved a dataset of approximately 55,000 Safe For Work (SFW) furry images sourced from e621.net, excluding ponies and scalies initially. These images were carefully cropped and aligned to focus on faces using a custom YOLOv3 network before training.
The AI model underwent extensive training for 33 days on a powerful TPUv3-32 pod, resulting in 3.2 million iterations to refine its image generation capabilities. To enhance the visual quality, the final generated images are scaled up to twice their original resolution and denoised using the Waifu2x algorithm. The creator emphasizes that as the images are AI-generated, they are non-copyrightable, allowing for free use, though responsible usage is encouraged.
Features
- AI Furry Portrait Generation: Utilizes nVidia's StyleGAN2 architecture to create unique furry faces.
- Large Training Dataset: Trained on ~55k SFW images from e621.net.
- Custom Face Detection: Used a custom YOLOv3 network for cropping and aligning faces in the training data.
- High-Resolution Output: Final images are upscaled (2x) and denoised using Waifu2x.
- Interactive Grid Display: Presents generated images in an interactive grid.
- User Controls: Offers controls for scroll speed, pause/resume, hover-zoom, fullscreen, panning, and opening images.
- Downloadable Model: Provides links to download the trained StyleGAN2 model.
Use Cases
- Generating inspiration for furry characters or art.
- Exploring the capabilities of AI image generation (StyleGAN2).
- Finding unique, non-copyrightable furry portraits for projects.
- Casual browsing and entertainment for furry enthusiasts.
FAQs
-
Are the generated images copyrighted?
No, as the images are generated by an AI, they are non-copyrightable. The creator claims no legal ownership or rights to any of the images generated by this AI. -
What AI model is used to generate the images?
The site uses nVidia's StyleGAN2 architecture, trained on a custom dataset. -
What was the training data for the AI?
The training dataset consisted of approximately 55,000 SFW furry images sourced from e621.net, cropped and aligned to faces.
Related Queries
Helpful for people in the following professions
Featured Tools
Join Our Newsletter
Stay updated with the latest AI tools, news, and offers by subscribing to our weekly newsletter.
