---
pretty_name: MOSS-Image Character and Style Catalog Samples
language:
- en
- ja
size_categories:
- 10K<n<100K
tags:
- image
- text-to-image
- synthetic
---

# MOSS-Image catalog data

**34,395 character samples + 314 styles = 34,709 generated images**, plus a 280-pixel thumbnail for each. Eight mutually exclusive style categories. Uncurated; no pass/fail labels. These are synthetic samples, not AnimaDex/Krea2 reference artwork. Not a validated character-recognition benchmark or an assertion of model training coverage.

[Open the catalog](https://huggingface.co/spaces/hide1/MOSS-Image-Character-Catalog) · [Project, generation settings and sample workflows](https://github.com/ukr8b3g-cmyk/MOSS-Image-Character-Catalog)

`catalog.json` holds public names, reusable prompts/tags, facets, source counts where available, timestamps and image paths/hashes. `images/` holds full-size WebP, `thumbnails/` derived previews. `style-categories.json` maps each style key to exactly one category. Files are divided by the first two ID characters.

EXIF/XMP removed without re-encoding full-size image payloads; private review state, credentials, local paths, workflows and generation logs are excluded. See [NOTICE.md](NOTICE.md) for rights and attribution scope. The MIT license applies to maintainer-authored viewer code/docs, not a blanket license to characters, artworks or generated images.

全件未選別です。正確さを保証するデータではありません。画像・プロンプトの再利用時はモデルの条件と第三者の権利を確認してください。生成設定とローカル版は上記GitHubで案内しています。

## Prompt wording revision — 2026-10-31

At the time of the 2026-10-31 revision, all 34,249 entries were checked with explicit text rules. Selected tied-hairstyle tags are written as descriptions of hair. Where the supplied appearance explicitly says android or robot, the contradictory human-species label is removed while gender is retained. Revised full-scene prompts replace the generic ears/tail/feet framing instruction with neutral whole-character framing. Open a card to copy the revised full-scene prompt.

Images have **not** been regenerated or curated. Original tags and saved workflows remain historical records; the revised text is for a new generation, not a claim about the prompt that produced the displayed image. Ambiguous source attributes still require review. No controlled generation test has established a reduction in unwanted tails.

The downloadable v0.1.2 ZIP also includes the prompt revision. On 2026-11-10, 432 new samples were added and one older actor-unspecified Luke was replaced; current total: 34,709.


## プロンプトの文章整理 — 2026-10-31

2026-10-31の修正時点では全34,249件を明示的な規則で点検しました。一部の結び髪タグを髪の英文説明に整理し、外見にandroid/robotが明記されている場合は矛盾する人間種の指定だけを除きました。性別の指定は維持しています。再生成用の全文では、耳・尻尾・足を列挙する共通構図文を中立な全身構図へ変更しました。カードを開くと修正版の再生成用プロンプトをコピーできます。

画像の再生成・選別はしていません。元タグと保存ワークフローは当時の記録です。修正文は今後の生成用で、表示画像が修正文から作られたという意味ではありません。元データの曖昧な特徴には確認が必要です。尻尾の誤生成が減ったかの同条件生成検証は未実施です。

今回の文章整理は更新済みv0.1.2 ZIPにも収録しています。2026-11-10に432枚を追加し、旧ルーク1枚を差し替え、現在の合計は34,709枚です。

## 2026-11-10 additions
432 new character samples (including two anime characters), one old Luke entry removed. Films / TV Series are independent filters. INT8 ConvRot SFT + LoRA 1.0, 8 steps, CFG 1.0, er_sde/simple, seed 8961184, 896×1184, WebP quality 70. Images remain uncurated; likeness and costume accuracy are not guaranteed.


## Classic Cartoon — 2026-11-10
29 new character samples (7 Flintstones + 22 others), 8 duplicates excluded. Independent Classic Cartoon work filter; reusable prompts explicitly include classic cartoon. SFT INT8 ConvRot + LoRA 1.0, 8 steps, CFG1, seed8961184, WebP70. Images remain uncurated. 現在34,395キャラクター＋314スタイル＝34,709枚。クラシック・カートゥーン分類を追加。閲覧専用デモです。