README.md

metadata

dataset_info:
  features:
    - name: modelId
      dtype: int64
    - name: modelName
      dtype: string
    - name: modelVersionId
      dtype: int64
    - name: modelVersionName
      dtype: string
    - name: fileId
      dtype: int64
    - name: fileName
      dtype: string
    - name: modelDescription
      dtype: string
    - name: allowNoCredit
      dtype: int64
    - name: allowCommercialUse
      dtype: string
    - name: allowDerivatives
      dtype: int64
    - name: allowDifferentLicense
      dtype: float64
    - name: modelType
      dtype: string
    - name: minor
      dtype: int64
    - name: poi
      dtype: int64
    - name: modelNsfwLevel
      dtype: int64
    - name: cosmetic
      dtype: float64
    - name: tags
      dtype: string
    - name: publishedAt
      dtype: string
    - name: baseModel
      dtype: string
    - name: baseModelType
      dtype: string
    - name: availability
      dtype: string
    - name: modelVersionNsfwLevel
      dtype: int64
    - name: modelVersionDescription
      dtype: string
    - name: trainedWords
      dtype: string
    - name: sizeKB
      dtype: float64
    - name: fileType
      dtype: string
    - name: fp
      dtype: string
    - name: format
      dtype: string
    - name: pickleScanResult
      dtype: string
    - name: pickleScanMessage
      dtype: string
    - name: virusScanResult
      dtype: string
    - name: scannedAt
      dtype: string
    - name: hashAutoV1
      dtype: string
    - name: hashAutoV2
      dtype: string
    - name: hashSHA256
      dtype: string
    - name: hashCRC32
      dtype: string
    - name: hashBLAKE3
      dtype: string
    - name: downloadUrl
      dtype: string
    - name: primary
      dtype: int64
    - name: downloadCount
      dtype: int64
    - name: ratingCount
      dtype: int64
    - name: rating
      dtype: float64
    - name: thumbsUpCount
      dtype: int64
    - name: thumbsDownCount
      dtype: int64
    - name: userHash
      dtype: string
  splits:
    - name: train
      num_bytes: 321605202
      num_examples: 155246
  download_size: 130378638
  dataset_size: 321605202
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

ImageGem Models

• project page: https://maps-research.github.io/imagegem-iccv2025/
• github repo with example usage: https://github.com/MAPS-research/imagegem
• ImageGem Images: https://huggingface.co/datasets/MAPS-research/ImageGem-images

Description

Name	Dtype	Source	Description
modelId	int64	model metadata
modelName	string	model metadata
modelVersionId	int64	model version metadata	each model may have multiple versions
modelVersionName	string	model version metadata
fileId	int64	model version file metadata	each model version may include multiple files (some may come with training data)
fileName	string	model version file metadata
modelDescription	string	model metadata
allowNoCredit	int64	model metadata	licencing from civitai, only 0 or 1
allowCommercialUse	string	model metadata	licencing from civitai, only 0 or 1
allowDerivatives	int64	model metadata	licencing from civitai, only 0 or 1
allowDifferentLicense	float64	model metadata	licencing from civitai, only 0 or 1
modelType	string	model metadata
minor	int64	model metadata
poi	int64	model metadata
modelNsfwLevel	int64	model metadata	the level rating from civitai
cosmetic	float64	model metadata
tags	string	model metadata	we aggregate each model's tags into a comma separated list
publishedAt	string	model version metadata
baseModel	string	model version metadata	the model structure that the LoRA is trained on
baseModelType	string	model version metadata
availability	string	model version metadata
modelVersionNsfwLevel	int64	model version metadata	the level rating from civitai
modelVersionDescription	string	model version metadata
trainedWords	string	model version metadata	A.K.A trigger words. We aggregate each model version's trained words into a comma separated list
sizeKB	float64	model version file metadata
fileType	string	model version file metadata
fp	string	model version file metadata
format	string	model version file metadata
pickleScanResult	string	model version file metadata
pickleScanMessage	string	model version file metadata
virusScanResult	string	model version file metadata
scannedAt	string	model version file metadata
hashAutoV1	string	model version file metadata
hashAutoV2	string	model version file metadata
hashSHA256	string	model version file metadata
hashCRC32	string	model version file metadata
hashBLAKE3	string	model version file metadata
downloadUrl	string	model version file metadata
primary	int64	model version file metadata	if this is the primary file for the model version, only 0 or 1
downloadCount	int64	model version file metadata
ratingCount	int64	model version stats
rating	float64	model version stats
thumbsUpCount	int64	model version stats
thumbsDownCount	int64	model version stats
userHash	string	model metadata	anonymized username

Notes

• All models in the table are LoRAs
• Each model on civitai may have multiple versions, and each version may have multiple files. Our table is a Cartesian product of models x versions x files
• We keep all available metadata from civitai api