100K Black and White Illustrations generated with FLUX.2 klein 4B

This dataset contains 94,188 black and white outline illustrations generated using the FLUX.2 klein 4B model by Black Forest Labs. The images follow a minimalistic ink-line art style on a pure white background, making them potentially useful for coloring book generation, clipart, icon datasets, or line-art style fine-tuning.

Generation Details

Style prompt applied:

Professional minimalistic black and white outline art. Pure white background, clean thick black ink lines, high contrast, vector style, no shading, no gradients, sharp edges.

Prompt Source

The text prompts used for generation were sourced from the Stable Diffusion 100k Custom Prompts and Images dataset by RS Turley on Kaggle, licensed under CC0: Public Domain. The original prompts were not modified.

Dataset Structure

dataset/
└── images/
    ├── metadata.csv                # text (prompt), file_name, seed — used by HF Dataset Viewer
    ├── generated_image_seed_*/
    │   ├── 1/
    │   ├── 2/
    │   └── ...

The metadata.csv file inside the images/ folder maps each image to its source prompt and generation seed, and is used by the HuggingFace Dataset Viewer to display labels alongside each image. Images are stored under the images/ folder, organized into numbered subfolders. Most subfolders contain around 1,000 images, but some may contain more due to a resumption bug during generation that caused the folder counter to reset, resulting in some folders accumulating more images than intended.

Download Script

Update: The Python scripts below may trigger Hugging Face API/HTTP rate limits. To avoid these errors and achieve faster downloads, it is recommended to clone the dataset using Git instead:

git clone https://huggingface.co/datasets/anokimchen/100k_BW_illustrations_i2i

A ready-to-use Python download script (download_dataset.py) is included in this repository. It lets you list all files, download everything, or download a specific range of images by index — useful if you only need a subset or want to download in batches.

A second script (1_download_and_resize.py) downloads images, resizes them to a target size (default 128×128), converts them to grayscale, and writes a local metadata CSV — useful for building training datasets.

Install the dependency:

pip install huggingface_hub

Examples:

# List all files without downloading
python download_dataset.py --list-only

# Download everything
python download_dataset.py

# Download images only (no metadata CSV)
python download_dataset.py --images-only

# Download metadata CSV only
python download_dataset.py --metadata-only

# Download a specific range of images by index (1-based, inclusive)
python download_dataset.py --range 1-100
python download_dataset.py --range 500-900
python download_dataset.py --range 1-100 500-900 2000-2500

# Download to a custom folder
python download_dataset.py --output-dir /path/to/my/folder

# Combine: download image range to a custom folder
python download_dataset.py --images-only --range 1-100 --output-dir ./subset

Notes:

• Index is 1-based and refers to the sorted list of image files in the repo.
• Ranges are inclusive on both ends (1-100 downloads images 1 through 100).
• Multiple ranges can be passed at once.
• Files that already exist locally are skipped automatically, so it is safe to resume a partial download.

⚠️ Image Quality Notice

While a black and white outline style was applied consistently via the prompt, not every image has been manually verified. Some images may:

• Not be truly black and white
• Contain color artifacts or shading
• Not fully conform to the intended minimalistic style

The dataset is provided as-is. Users should verify images independently before use in any downstream task. Use at your own risk.

⚠️ Disclaimer

This dataset is provided as-is for research and educational purposes. The images are AI-generated and were produced automatically from third-party prompts without manual review of each output.

The dataset creator:

• Makes no guarantees about the suitability, accuracy, safety, or fitness of these images for any particular purpose
• Is not responsible for any use, misuse, or downstream application of this dataset or its contents by third parties

Users are solely responsible for ensuring their use of this dataset complies with applicable laws and regulations in their jurisdiction.

Some prompts from the source dataset reference fictional characters, public figures, or brand names — users should exercise their own judgment regarding any legal or ethical considerations that may apply to specific images.

License

• Images: Generated using FLUX.2 klein 4B under the Apache 2.0 license
• Source prompts: CC0 Public Domain (via Kaggle dataset by RS Turley)
• Download script: CC0 Public Domain — free to use, modify, and redistribute without restriction