datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
PIPE
Dataset Card for PIPE Dataset
Dataset Summary
The PIPE (Paint by InPaint Edit) dataset is designed to enhance the efficacy of mask-free, instruction-following image editing models by providing a large-scale collection of image pairs and diverse object addition instructions. Comprising approximately 1 million image pairs, PIPE includes both source and target images, along with corresponding natural language instructions for object addition. The dataset leverages extensive… See the full description on the dataset page: https://huggingface.co/datasets/paint-by-inpaint/PIPE.chinese-painting-collection
Chinese Painting Collection
91,438 images of Chinese paintings with bilingual (Chinese/English) VLM captions,
calligraphy OCR transcriptions, view-type classification, a long caption on part
of the set, and foreign-object detection with one final crop box per image.
Sources
Component
Source
Images
Image license
npm_tw_c0–npm_tw_c3
National Palace Museum (Taipei) Open Data
86,666
Taiwan Open Government Data License v1 (attribution required)
met_china… See the full description on the dataset page: https://huggingface.co/datasets/kaupane/chinese-painting-collection.Chinese_Landscape_Painting
数据集名称
Chinese_Landscape_Painting
数据集简介
这是一份为南京大学智能科学与技术专业大二秋季学期课程人工智能导论课程的项目训练而搭建的数据集。
由于目前较大规模、高质量、且适应现代flux模型的高分辨率的山水画数据集稀缺，故我们搭建了这个数据集，以进行flux模型的lora微调训练。
数据集包含了1017张局部图片，79张全景图片，全部采样自中国山水画的十大名画，并全部带有精细的严格结构化标注。
如果你想使用该数据集进行flux模型训练，可以直接下载并自行修改相应的子文件夹名称以改变每张图片的训练次数。
引用说明
如果您在研究或项目中使用了本数据集，请按以下格式引用：
BibTeX:
@dataset{Chinese_Landscape_Painting,
author = {Wei Liangxu},
title = {Chinese_Landscape_Painting},
year = {2025}… See the full description on the dataset page: https://huggingface.co/datasets/AmazarashiEndure/Chinese_Landscape_Painting.chinese_landscape_paintings
Dataset Card for "chinese_landscape_paintings"
More Information needed
painting_256painting_movements_2painting_movements3pixelsPIPE_Masks
Dataset Card for PIPE Masks Dataset
Dataset Summary
The PIPE (Paint by InPaint Edit) dataset is designed to enhance the efficacy of mask-free, instruction-following image editing models by providing a large-scale collection of image pairs and diverse object addition instructions.
Here, we provide the masks used for the inpainting process to generate the source image for the PIPE dataset for both the train and test sets.
Further details can be found in our project page… See the full description on the dataset page: https://huggingface.co/datasets/paint-by-inpaint/PIPE_Masks.painting-style-classification
Dataset Labels
['Realism', 'Art_Nouveau_Modern', 'Analytical_Cubism', 'Cubism', 'Expressionism', 'Action_painting', 'Synthetic_Cubism', 'Symbolism', 'Ukiyo_e', 'Naive_Art_Primitivism', 'Post_Impressionism', 'Impressionism', 'Fauvism', 'Rococo', 'Minimalism', 'Mannerism_Late_Renaissance', 'Color_Field_Painting', 'High_Renaissance', 'Romanticism', 'Pop_Art', 'Contemporary_Realism', 'Baroque', 'New_Realism', 'Pointillism', 'Northern_Renaissance', 'Early_Renaissance'… See the full description on the dataset page: https://huggingface.co/datasets/keremberke/painting-style-classification.Sumi-e_no_kurozuappu_ink_painting_closeups
Sumi-e no Kurozuappu Ink Painting Closeups
Dataset Description
This dataset contains 1,250 synthetically generated images of Japanese Sumi-e ink paintings. The images were created to explore the capabilities of diffusion models in replicating the nuanced art style of traditional Japanese ink painting. The dataset is ideal for those interested in niche artistic model creation, generating LORAs, or simply enjoying and sharing these artistic expressions.
Example… See the full description on the dataset page: https://huggingface.co/datasets/takara-ai/Sumi-e_no_kurozuappu_ink_painting_closeups.genshin_impact_CHONGYUN_Paints_UNDO_Sketch_UnChunked
northern-landscape-painting
License
This dataset was compiled for the purpose of art historical research on Scandinavian landscape paintings.
Images scraped from Wikimedia Commons via Wikidata; metadata scraped from
Wikidata (CC0). Image licenses vary per file (predominantly public domain,
some CC-BY-SA). See the license_short_name / license_url columns in
the parquet files for the exact terms of each individual image, and the
Commons file page for full details.
PaintBench
PaintBench
Deterministic Evaluation of Precise Visual Editing
Kai Xu*,
Ellis Brown*,
Shrikar Madhu,
Rob Fergus,
He He,
Saining Xie
*Equal contribution.
New York University
A precise, deterministic visual editing benchmark. Evaluates whether native pixel-space image generation models can execute "MS-Paint-style" edits — geometric transforms, color changes, structural manipulation, and symbolic-reasoning edits — with pixel-level… See the full description on the dataset page: https://huggingface.co/datasets/PaintBench/PaintBench.PaintBench
PaintBench
A precise, deterministic visual editing benchmark. PaintBench evaluates whether generative models can execute MS-Paint-style image editing operations with pixel-level correctness. Every problem is a (input_image, instruction, answer_image) triplet, generated programmatically so the answer is pixel-exact and the answer distribution is known by construction.
This dataset bundles three sibling benchmarks that share one generation and evaluation pipeline:
Subset… See the full description on the dataset page: https://huggingface.co/datasets/PaintBenchAnonymousNeurIPS26/PaintBench.finetuning-landscape-paintingThis dataset was compiled for the purpose of finetuning models in the context of benchmarking for art historical research.
Images scraped from Wikimedia Commons via Wikidata; metadata scraped from
Wikidata (CC0). Image licenses vary per file (predominantly public domain,
some CC-BY-SA) see the license_short_name / license_url columns in the
parquet files for the exact terms of each individual image, and the
commons file page for full details.
dirty-painting-a3c0ac
dirty-painting-a3c0ac
Synthetic sensors test data: 53 rows in data.csv.
All values are randomly generated fictional examples, not real observations, products, or user activity. Intended only for CSV loading and pipeline tests; not suitable for scientific or business conclusions. Columns are sampled independently and do not model real-world correlations.
Fields
sample_id: random identifier for this generated sample.
row_id: sequential row number starting at 1.… See the full description on the dataset page: https://huggingface.co/datasets/Harbor-James/dirty-painting-a3c0ac.stupid-paint-aaed43
stupid-paint-aaed43
Synthetic weather test data: 36 rows in data.csv.
All values are randomly generated fictional examples, not real observations, products, or user activity. Intended only for CSV loading and pipeline tests; not suitable for scientific or business conclusions. Columns are sampled independently and do not model real-world correlations.
Fields
sample_id: random identifier for this generated sample.
row_id: sequential row number starting at 1.… See the full description on the dataset page: https://huggingface.co/datasets/charles-harris/stupid-paint-aaed43.kr_ink_painting_05kr_ink_painting_07chinese_landscape_paintings_1k
Dataset Card for "chinese_landscape_paintings_1k"
More Information needed
Stylized-Painterly-Textures-v1.1This is a dataset consisting of 19 images for finetuning image models to generate textures in a black and white, painterly style that can be used in 3D.
v1.1 improvements:
- All the images are now completely seamless/tileable
- Better rock texture
- Better metal grate texture
Example images from this dataset:
How this dataset was created:
I generated images using un-finetuned Qwen Image using the below prompt, cherry picking only the best results until I had 10 imagesFinetuned Qwen… See the full description on the dataset page: https://huggingface.co/datasets/Hyperccino/Stylized-Painterly-Textures-v1.1.wan_paint_over_effectThis dataset contains videos generated using Wan 2.1 T2V 14B.
artinsight-painting-deterioration
ArtInsight — Easel Painting Deterioration Detection
20 high-resolution full-frame photographs of easel paintings, annotated by expert
restorers with the areas of damage they see. 2,909 annotations in total.
Conservation assessment is normally done by eye, by specialists, one painting at a time.
This dataset is an attempt to make it learnable.
Two damage types, two disjoint sets of paintings
Subset
Images
What it marks
LPL
8
Lacuna from Loss of the… See the full description on the dataset page: https://huggingface.co/datasets/biglam/artinsight-painting-deterioration.autotrain-kalighat-paintings-lorakr_ink_painting_08genshin_impact_ALBEDO_hidiffusion_oil_painting_and_texts
part3-paint_clean_brushThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "hand",
"total_episodes": 1894,
"total_frames": 968485,
"total_tasks": 18,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:1894"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/gsethia08/part3-paint_clean_brush.Stylized-Painterly-Textures-v1.0This is a dataset consisting of 19 images for finetuning image models to generate textures in a black and white, painterly style that can be used in 3D.
Example images from this dataset:
How this dataset was created:
I generated images using un-finetuned Qwen Image using the below prompt, cherry picking only the best results until I had 10 images
Finetuned Qwen Image on those 10 images
Used that finetune to re-generate better versions of the least good of the original 10 images +… See the full description on the dataset page: https://huggingface.co/datasets/Hyperccino/Stylized-Painterly-Textures-v1.0.oil_painting_dataset