CoolFace
Datasetpublic

huanx/GalMaster-style-lora-dataset

GalMaster Style LoRA Dataset Overview 329 images (145 CG + 184 character illustrations) Source: GalMaster (Galgame) Purpose: Style LoRA training for Stable Diffusion Structure All files are flat in the repo root (no subdirectories): 329 image files (.png / .jpg) 329 caption files (.txt, same name as image) tag_summary.csv — caption summary for all 329 images Captioning Tagger: WD ConvNeXt Tagger v3… See the full description on the dataset page: https://huggingface.co/datasets/huanx/GalMaster-style-lora-dataset.

sourceHugging Faceotherupdated 2mo agoView on Hugging Face
0likes172downloads
Dataset Card

GalMaster Style LoRA Dataset

Overview

  • —329 images (145 CG + 184 character illustrations)
  • —Source: GalMaster (Galgame)
  • —Purpose: Style LoRA training for Stable Diffusion

Structure

All files are flat in the repo root (no subdirectories):

  • —329 image files (.png / .jpg)
  • —329 caption files (.txt, same name as image)
  • —tag_summary.csv — caption summary for all 329 images

Captioning

  • —Tagger: WD ConvNeXt Tagger v3 (SmilingWolf/wd-convnext-tagger-v3)
  • —Threshold: general=0.35, character=0.85
  • —Style prefix: game_cg, galgame_style
  • —Character tags (category=4): 0 leaked
  • —Dropped false-positive tags: virtualyoutuber, abyssalship, healthbar, gameplaymechanics, bluetongue, coloredtongue, wavymouth, noticelines
  • —Each image has a corresponding .txt caption file with the same name

Character mapping (03_立绘)

CodeNamePinyin
MM晓夜Xiāo Yè
BXY白纤语Bái Xiān Yǔ
MCL莫楚玲Mò Chǔ Líng
HXC韩小澄Hán Xiǎo Chéng

Code suffix 1/2 = two outfits per character (uniform/casual).

Stats

  • —Unique tags: 247
  • —Top tags: 1girl(316), longhair(295), solo(288), coloredskin(258), longsleeves(236), skirt(235), shirt(232), simplebackground(212), hairornament(211), whitebackground(203)
  • —Tag count per caption: min=6, max=42, mean=25.4

Tools

  • —Tagger script: tag.py (WD ConvNeXt v3 ONNX, raw [0,255] input, NHWC)
  • —Model: SmilingWolf/wd-convnext-tagger-v3