OpenBrush Anonymous Masters: 41,914 Unattributed Artworks for Style Training

41,914 unattributed artworks form a curated subset of the OpenBrush-75K dataset, designed to provide broad-style training data without artist-specific bias. The dataset was created by jaddai and was last updated on 2026-05-27. It is derived from a parent collection of 75,313 images, with the same CC0 license and caption schema.

Use Cases

• Train generative image models on diverse artistic styles based on the broad collection of unattributed works.
• Fine-tune vision-language models using the consistent caption schema mentioned in the description.
• Conduct art history or style analysis research using a large corpus of anonymous masterworks.
• Benchmark style transfer algorithms on a dataset free from known artist attribution.

Strengths

• Contains 41,914 images, a substantial curated subset focused on unattributed works.
• Derived from a larger parent dataset of 75,313 images, indicating a defined source.
• Explicitly designed to remove artist-specific bias for broad-style training.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full parent dataset is known, but specific file formats and size for this subset are unknown.
• Data may reflect temporal and stylistic bias inherent to the source collection of 'old masters'.

Provenance

Source

Curated subset of jaddai/openbrush-75k on Hugging Face.

Collection Method

Filtered from a larger collection using a VLM (Qwen3-VL-30B-A3B) for captioning, selecting unattributed works.

Time Range

Covers centuries of art history, but specific temporal coverage is not defined.

Freshness

Last updated 2026-05-27 21:24:00; freshness should be verified.