---
license: apache-2.0
---

Most of these (at least 70% total, almost all difficult/complex scenes) are human-reviewed and human-edited, based on Gemini 3 Flash captions with a pipeline following multi-crop passes and combination.

Images (plus regularization images and other uncaptioned images): https://huggingface.co/datasets/furproxy/img_dump

Each file can have multiple lines. Each text file is an image caption. Images are sourced from e621/civit/booru, stored as md5 name.

Currently ~7.5k captions of a furry dataset with the general format:

- Scene introduction and characters
- Interactions and visible character details
- Background or other elements and style

Please do feel free to let me know if there are issues or if you want to contribute. This did take a lot of time, effort, and money.

Note: There are small occasional issues that got through describing "looking over shoulder"/"behind"/"in front of", because it can be ambiguous.