---
title: SakuraPool
canonical_url: "https://www.modelscope.cn/datasets/leafmoone/SakuraPool"
md_url: "https://www.modelscope.cn/datasets/leafmoone/SakuraPool.md"
repository: leafmoone/SakuraPool
last_updated: 2026-10-09
license: apache-2.0
storage_size: "13 GB"
downloads: 1732
stars: 1
---

# SakuraPool

> SakuraPool - leafmoone 在 ModelScope 开源的数据集。Seven source-separated SakuraPool index publications; no original images

leafmoone/SakuraPool 是 ModelScope 魔搭社区上的数据集，存储大小 13 GB，采用 apache-2.0 许可。

- **Repository**: leafmoone/SakuraPool
- **License**: apache-2.0
- **Storage size**: 13 GB
- **Downloads**: 1732
- **Stars**: 1
- **Last updated**: 2026-10-09

Source: https://www.modelscope.cn/datasets/leafmoone/SakuraPool

---

# SakuraPool source-separated index publications

**Index metadata only. No original image TARs or image files are distributed here.**

This fixed release covers **715 committed partitions, 22,792 indexed TAR objects,
and 26,430,011 records**. Counts were verified from committed Parquet fragments;
all COMMIT/fragment size, SHA-256, schema, row count, and ownership checks passed.
The indexed TAR path set and sizes/SHA-256 exactly match the fixed provider metadata.
These counts are physical retained records, not a claim of unique image content.

Each source is an independent complete local SakuraPool publication with its own
runtime catalog, bitmaps, locations, image hashes, immutable snapshot, and remote
object catalog. Do not mix files or snapshots between sources. All records in a
publication belong to its named source. There are 12 required files per source.

| Source | Partitions | TAR objects | Records | Installed index bytes |
|---|---:|---:|---:|---:|
| anime_pictures | 32 | 1,000 | 642,954 | 169,066,240 |
| bangumi | 94 | 3,000 | 4,168,774 | 836,745,004 |
| danbooru | 361 | 11,552 | 11,132,809 | 2,978,602,701 |
| gamecg | 8 | 240 | 5,177,447 | 1,212,210,257 |
| konachan | 32 | 1,000 | 318,760 | 80,221,647 |
| yande | 94 | 3,000 | 1,124,564 | 245,637,771 |
| zerochan | 94 | 3,000 | 3,864,703 | 935,541,500 |

## Versions and provenance


Private management manifests, build INPUTs, absolute machine paths, credentials,
signed URLs, logs, P2 archive payloads and original images are not mirrored.
Runtime OWNER/STAGE markers are necessary format files, not checkpoint admin data.
`catalog.json` lists the exact required relative paths and immutable publication IDs.

## Index access and local use

Download **all** files listed for your source in `catalog.json` from one immutable
repository revision, preserving paths. The local publication root is
`LOCAL_REPO/sources/konachan` (or another listed source), not the repository root.
There is no automatic index discovery/installer in SakuraPool; index acquisition
is explicit and separate from image retrieval. With an installed SakuraPool:

```text
sakura publication verify LOCAL_REPO/sources/konachan --full
```

Put `{"sources":["konachan"]}` in `source-query.json`, then create a small task:

```text
sakura task create --publication LOCAL_REPO/sources/konachan --query source-query.json --selection first --limit 3 --task-dir TASK
sakura task run TASK --profile TASK_PROFILE.json
```

Task creation is local and does not read credentials or make HTTP requests.
`publication fetch` retrieves selected image records; it does not install indices.

## Original-image access and rights



Apache-2.0 applies to the distributed index metadata, consistent with upstream
metadata licensing. It is **not** a license grant for original images, illustrations,
characters, third-party metadata, privacy or other underlying rights. Original
rights remain with their respective owners; comply with upstream access terms and
applicable law. This repository provides no image redistribution authorization.