README.md

---
dataset_info:
  features:
  - name: approver_id
    dtype: float64
  - name: bit_flags
    dtype: int64
  - name: created_at
    dtype: string
  - name: down_score
    dtype: int64
  - name: fav_count
    dtype: int64
  - name: file_ext
    dtype: string
  - name: file_size
    dtype: int64
  - name: file_url
    dtype: string
  - name: has_active_children
    dtype: bool
  - name: has_children
    dtype: bool
  - name: has_large
    dtype: bool
  - name: has_visible_children
    dtype: bool
  - name: id
    dtype: int64
  - name: image_height
    dtype: int64
  - name: image_width
    dtype: int64
  - name: is_banned
    dtype: bool
  - name: is_deleted
    dtype: bool
  - name: is_flagged
    dtype: bool
  - name: is_pending
    dtype: bool
  - name: large_file_url
    dtype: string
  - name: last_comment_bumped_at
    dtype: string
  - name: last_commented_at
    dtype: string
  - name: last_noted_at
    dtype: string
  - name: md5
    dtype: string
  - name: media_asset_created_at
    dtype: string
  - name: media_asset_duration
    dtype: float64
  - name: media_asset_file_ext
    dtype: string
  - name: media_asset_file_key
    dtype: string
  - name: media_asset_file_size
    dtype: int64
  - name: media_asset_id
    dtype: int64
  - name: media_asset_image_height
    dtype: int64
  - name: media_asset_image_width
    dtype: int64
  - name: media_asset_is_public
    dtype: bool
  - name: media_asset_md5
    dtype: string
  - name: media_asset_pixel_hash
    dtype: string
  - name: media_asset_status
    dtype: string
  - name: media_asset_updated_at
    dtype: string
  - name: media_asset_variants
    dtype: string
  - name: parent_id
    dtype: float64
  - name: pixiv_id
    dtype: float64
  - name: preview_file_url
    dtype: string
  - name: rating
    dtype: string
  - name: score
    dtype: int64
  - name: source
    dtype: string
  - name: tag_count
    dtype: int64
  - name: tag_count_artist
    dtype: int64
  - name: tag_count_character
    dtype: int64
  - name: tag_count_copyright
    dtype: int64
  - name: tag_count_general
    dtype: int64
  - name: tag_count_meta
    dtype: int64
  - name: tag_string
    dtype: string
  - name: tag_string_artist
    dtype: string
  - name: tag_string_character
    dtype: string
  - name: tag_string_copyright
    dtype: string
  - name: tag_string_general
    dtype: string
  - name: tag_string_meta
    dtype: string
  - name: up_score
    dtype: int64
  - name: updated_at
    dtype: string
  - name: uploader_id
    dtype: int64
  splits:
  - name: train
    num_bytes: 20051410186
    num_examples: 8616173
  download_size: 7310216883
  dataset_size: 20051410186
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: mit
task_categories:
- text-to-image
- image-classification
language:
- en
- ja
pretty_name: Danbooru 2025 Metadata
size_categories:
- 1M<n<10M
---

# Dataset Card for Danbooru 2025 Metadata

This dataset repo provides comprehensive, up-to-date metadata for the Danbooru booru site. All metadata was freshly scraped starting on **January 2, 2025**, resulting in more extensive tag annotations for older posts, fewer errors, and reduced occurrences of non-labelled AI-generated images in the data.

## Dataset Details

**What is this?**  
A refreshed, Parquet-formatted metadata dump of Danbooru, current as of January 2, 2025.

**Why this over other Danbooru scrapes?**  
- **Fresh Metadata:** Coverage includes post IDs from 1 through ~8.6M, with the newest vocabulary and tag annotations.  
- **Maximized Tag Count:** Many historical tag renames and additions are accurately reflected, reducing duplications for downstream tasks.  
- **Reduced Noise:** Fewer untagged or mislabeled AI images compared to older scrapes.

**Tag Comparisons**  
[TODO: Contrast the tag counts, deleted entries, etc. with other Danbooru metadata scrapes.]

- **Shared by:** [trojblue](https://huggingface.co./trojblue)  
- **Language(s) (NLP):** English, Japanese  
- **License:** MIT  

## Uses

The dataset can be loaded or filtered with the Huggingface `datasets` library:

```python
from datasets import Dataset, load_dataset

danbooru_dataset = load_dataset("trojblue/danbooru2025-metadata", split="train")
df = danbooru_dataset.to_pandas()
```

This dataset can be used to:
- Retrieve the full Danbooru image set via the metadata’s URLs
- Train or fine-tune an image tagger
- Compare against previous metadata versions to track changes, tag evolution, and historical trends


## Dataset Structure

Below is a partial overview of the DataFrame columns, derived directly from the Danbooru JSONs:


```python
import unibox as ub
ub.peeks(df)
```

```
(8616173, 59)
Index(['approver_id', 'bit_flags', 'created_at', 'down_score', 'fav_count',
       'file_ext', 'file_size', 'file_url', 'has_active_children',
       'has_children', 'has_large', 'has_visible_children', 'id',
       'image_height', 'image_width', 'is_banned', 'is_deleted', 'is_flagged',
       'is_pending', 'large_file_url', 'last_comment_bumped_at',
       'last_commented_at', 'last_noted_at', 'md5', 'media_asset_created_at',
       'media_asset_duration', 'media_asset_file_ext', 'media_asset_file_key',
       'media_asset_file_size', 'media_asset_id', 'media_asset_image_height',
       'media_asset_image_width', 'media_asset_is_public', 'media_asset_md5',
       'media_asset_pixel_hash', 'media_asset_status',
       'media_asset_updated_at', 'media_asset_variants', 'parent_id',
       'pixiv_id', 'preview_file_url', 'rating', 'score', 'source',
       'tag_count', 'tag_count_artist', 'tag_count_character',
       'tag_count_copyright', 'tag_count_general', 'tag_count_meta',
       'tag_string', 'tag_string_artist', 'tag_string_character',
       'tag_string_copyright', 'tag_string_general', 'tag_string_meta',
       'up_score', 'updated_at', 'uploader_id'],
      dtype='object')
```

<div>
<style scoped>
    .dataframe tbody tr th:only-of-type {
        vertical-align: middle;
    }

    .dataframe tbody tr th {
        vertical-align: top;
    }

    .dataframe thead th {
        text-align: right;
    }
</style>
<table border="1" class="dataframe">
  <thead>
    <tr style="text-align: right;">
      <th></th>
      <th>approver_id</th>
      <th>bit_flags</th>
      <th>created_at</th>
      <th>down_score</th>
      <th>fav_count</th>
      <th>file_ext</th>
      <th>file_size</th>
      <th>file_url</th>
      <th>has_active_children</th>
      <th>has_children</th>
      <th>...</th>
      <th>tag_count_meta</th>
      <th>tag_string</th>
      <th>tag_string_artist</th>
      <th>tag_string_character</th>
      <th>tag_string_copyright</th>
      <th>tag_string_general</th>
      <th>tag_string_meta</th>
      <th>up_score</th>
      <th>updated_at</th>
      <th>uploader_id</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <th>0</th>
      <td>NaN</td>
      <td>0</td>
      <td>2015-08-07T23:23:45.072-04:00</td>
      <td>0</td>
      <td>66</td>
      <td>jpg</td>
      <td>4134797</td>
      <td>https://cdn.donmai.us/original/a1/b3/a1b3d0fa9...</td>
      <td>False</td>
      <td>False</td>
      <td>...</td>
      <td>3</td>
      <td>1girl absurdres ass bangle bikini black_bikini...</td>
      <td>kyouka.</td>
      <td>marie_(splatoon)</td>
      <td>splatoon_(series) splatoon_1</td>
      <td>1girl ass bangle bikini black_bikini blush bra...</td>
      <td>absurdres commentary_request highres</td>
      <td>15</td>
      <td>2024-06-25T15:32:44.291-04:00</td>
      <td>420773</td>
    </tr>
    <tr>
      <th>1</th>
      <td>NaN</td>
      <td>0</td>
      <td>2008-03-05T01:52:28.194-05:00</td>
      <td>0</td>
      <td>7</td>
      <td>jpg</td>
      <td>380323</td>
      <td>https://cdn.donmai.us/original/d6/10/d6107a13b...</td>
      <td>False</td>
      <td>False</td>
      <td>...</td>
      <td>2</td>
      <td>1girl aqua_hair bad_id bad_pixiv_id guitar hat...</td>
      <td>shimeko</td>
      <td>hatsune_miku</td>
      <td>vocaloid</td>
      <td>1girl aqua_hair guitar instrument long_hair so...</td>
      <td>bad_id bad_pixiv_id</td>
      <td>4</td>
      <td>2018-01-23T00:32:10.080-05:00</td>
      <td>1309</td>
    </tr>
    <tr>
      <th>2</th>
      <td>85307.0</td>
      <td>0</td>
      <td>2015-08-07T23:26:12.355-04:00</td>
      <td>0</td>
      <td>10</td>
      <td>jpg</td>
      <td>208409</td>
      <td>https://cdn.donmai.us/original/a1/2c/a12ce629f...</td>
      <td>False</td>
      <td>False</td>
      <td>...</td>
      <td>1</td>
      <td>1boy 1girl blush boots carrying closed_eyes co...</td>
      <td>yuuryuu_nagare</td>
      <td>jon_(pixiv_fantasia_iii) race_(pixiv_fantasia)</td>
      <td>pixiv_fantasia pixiv_fantasia_3</td>
      <td>1boy 1girl blush boots carrying closed_eyes da...</td>
      <td>commentary_request</td>
      <td>3</td>
      <td>2022-05-25T02:26:06.588-04:00</td>
      <td>95963</td>
    </tr>
  </tbody>
</table>
</div>


## Dataset Creation

We scraped all post IDs on Danbooru from 1 up to the latest. Some restricted tags (e.g. `loli`) were hidden by the site and require a gold account to access, so they are not present.  
For a more complete (but older) metadata reference, you may wish to combine this with Danbooru2021 or similar previous scrapes.

The scraping process used a pool of roughly 400 IPs over six hours, ensuring consistent tag definitions. Below is a simplified example of the process used to convert the metadata into Parquet:

```python
import pandas as pd
from pandarallel import pandarallel

# Initialize pandarallel
pandarallel.initialize(nb_workers=4, progress_bar=True)

def flatten_dict(d, parent_key='', sep='_'):
    """
    Flattens a nested dictionary.
    """
    items = []
    for k, v in d.items():
        new_key = f"{parent_key}{sep}{k}" if parent_key else k
        if isinstance(v, dict):
            items.extend(flatten_dict(v, new_key, sep=sep).items())
        elif isinstance(v, list):
            items.append((new_key, ', '.join(map(str, v))))
        else:
            items.append((new_key, v))
    return dict(items)

def extract_all_illust_info(json_content):
    """
    Parses and flattens Danbooru JSON into a pandas Series.
    """
    flattened_data = flatten_dict(json_content)
    return pd.Series(flattened_data)

def dicts_to_dataframe_parallel(dicts):
    """
    Converts a list of dicts to a flattened DataFrame using pandarallel.
    """
    df = pd.DataFrame(dicts)
    flattened_df = df.parallel_apply(lambda row: extract_all_illust_info(row.to_dict()), axis=1)
    return flattened_df
```


### Recommendations

Users should be aware of potential biases and limitations, including the presence of adult content in some tags. More details and mitigations may be needed.