| # GenImage demonstration images | |
| Eight original files from the **GenImage validation split**: four real ImageNet | |
| images (`nature`, label 0) and four generated images (`ai`, label 1), covering | |
| Stable Diffusion 1.4, Midjourney, ADM and BigGAN. | |
| Source: [GenImage official repository](https://github.com/GenImage-Dataset/GenImage). | |
| `manifest.json` records each original dataset-relative path, dimensions, label | |
| and SHA-256. Files are copied byte for byte; thumbnails are only UI previews. | |
| Samples were selected for visual variety before running model predictions. | |
| This small gallery is not a benchmark evaluation. | |
| Image rights remain with the original datasets and rightsholders. The software's | |
| MIT license does not relicense these image assets. | |