---
license: apache-2.0
task_categories:
- text-to-image
- image-to-text
pretty_name: Civitai Images
size_categories:
- 1M<n<10M
tags:
- stable diffusion
- prompts
---

# Civitai Images

Images+metadata from [Civitai](https://civitai.com/)

Stats:
- ~4.1M

Formats:
- WebDataset
    - 10k per shard, ~2GB
    - jpg + json
    - `__key__` is Civitai image id

## Notes
- ~464k images with no `meta` field are excluded, this is ~10% of images collected
- Files for some entries are actually videos, these will be released separately
- Civitai [extract metadata](https://github.com/hlky/OpenCivit/tree/main/src/utils/metadata) on upload, the exact fields in `meta` will depend on the UI used, some are common e.g. `prompt`, others are UI specific
- Includes reaction data

## another BIG data banger straight from the underground

## with thanks to Civitai and their community ❤️