CoolFace
Datasetpublic

data-archetype/LAION_Aesthetics_1024_bucketed_1024

LAION Aesthetics 1024 Bucketed 1024 Captioned Bucketed-shards export of images from limingcv/LAION_Aesthetics_1024, with a small high-resolution addon from limingcv/LAION_Aesthetics_512. Images: 382,412 Shards: 398 uncompressed WebDataset-style TAR files Format: bucketed_shards_v1 Base resolutions: [1024, 512] Manifest: manifest.json Images are organized under buckets/<bucket_id>/<shard>.tar. Each sample contains: <key>.jpg <key>.txt <key>.json The .txt files contain… See the full description on the dataset page: https://huggingface.co/datasets/data-archetype/LAION_Aesthetics_1024_bucketed_1024.

sourceHugging Faceotherupdated 2mo agoView on Hugging Face
0likes643downloads
Dataset Card

LAION Aesthetics 1024 Bucketed 1024 Captioned

Bucketed-shards export of images from limingcv/LAION_Aesthetics_1024, with a small high-resolution addon from limingcv/LAION_Aesthetics_512.

  • —Images: 382,412
  • —Shards: 398 uncompressed WebDataset-style TAR files
  • —Format: bucketed_shards_v1
  • —Base resolutions: [1024, 512]
  • —Manifest: manifest.json

Images are organized under buckets/<bucket_id>/<shard>.tar. Each sample contains:

  • —<key>.jpg
  • —<key>.txt
  • —<key>.json

The .txt files contain model-generated captions. The original LAION scrape text is not used as the training caption.

Caption priority:

  1. 1.google/gemini-2.5-flash-lite: 32,941
  2. 2.google/gemini-2.0-flash-001: 348,861
  3. 3.mistralai/mistral-medium-3.1: 147
  4. 4.openai/gpt-5-mini: 463

Most images were resized/cropped into SDXL-style 1024 buckets without upsampling. The 463-image addon is stored as passthrough JPEG bytes and assigned to matching 1024 buckets; loaders should use the bucket metadata in each sample JSON/manifest for target dimensions.

See manifest.json for bucket counts, shard checksums, caption source IDs, and prompt hashes. Prompt text is intentionally not included.