Beach Does Not Exist

Are the beaches generated by this tool real or AI-generated?

All beaches are AI-generated. The tool uses a StyleGAN2-ADA network trained on a dataset of beach images to create synthetic beach visuals at 1280×768 pixels.

How are AI-generated beach images created on this tool?

The tool employs a StyleGAN2-ADA network trained on approximately 20,000 beach images. Images are generated at a resolution of 1280×768 pixels, and all beaches produced are synthetic.

Can I download the model/network or use it for transfer learning?

Yes. You can download the network package (362 MB) for your experiments or retrain the network with your own dataset via transfer learning.

What are the key technical details of the model?

• Architecture: StyleGAN2-Ada
• Dataset: ~20,000 beach images
• Network size: 362 MB
• Resolution: 1280×768 px
• Network layout: 5×3 squares (with 28 = 256 px sides)
• Training options: --min_w=5 --min_h=3 --res\_log2=8
• Training steps: 25,000 kimg
• FID50k: 3.14
• Original framework: StyleGAN2-ADA (NVIDIA research lineage)
• Author: Vojtěch Semecký
• Contact: info@thisbeachdoesnotexist.com

How is training progress tracked and what does 25,000 kimg mean?

Training progress is shown as kimg (kilo images). The network was trained for 25,000 kimg, which corresponds to about 25,000,000 images seen during training.

What is the FID score for this model?

The model reports an FID50k score of 3.14.

What is K-Means Clustering used for in this tool?

K-Means clustering groups similar AI-generated beach images to help users explore variations, segment images into clusters based on similarities, and identify common features among random images.

Who maintains the site and how is it monitored?

The site is monitored by Testomato.com.