datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
conceptual-captions-12m-webdataset-metadata
Conceptual Captions 12M — Webshart metadata indices
Per-shard webshart metadata indices for
laion/conceptual-captions-12m-webdataset:
1,100 JSON files under data/, one per source tar shard, mirroring the source's shard layout.
Each index records every tar member's byte offset and length (enabling ranged reads without
downloading whole shards), image geometry (width/height for aspect bucketing), and — as of
August 2026 — embedded captions for all 10,994,853 samples, coalesced… See the full description on the dataset page: https://huggingface.co/datasets/webshart/conceptual-captions-12m-webdataset-metadata.suno-various-94k
Suno Various 94k: originals, ACE-Step covers, and stems
A unified, indexed repository containing 94,174 exact source/cover pairs for
reference conditioning, preference, slider, source-separation, and music
style-transfer research. Four independently loadable configurations provide the
original tracks, generated covers, and four-source stems for both sides.
Configurations
Configuration
Shards
Samples
Contents
original
85
94,174
Original MP3 plus JSON… See the full description on the dataset page: https://huggingface.co/datasets/webshart/suno-various-94k.terminusresearch-photo-aesthetics
Photo Aesthetics — Webshart
This is the canonical, maintained location for the former terminusresearch/photo-aesthetics dataset. The migration completed on August 19, 2026, with 30,032 image samples repackaged into 371 indexed Webshart shards. Every sample now includes both the original CogVLM caption and a newer GLM-5.3-Flash caption.
The legacy repository now contains only a relocation notice. Its original tar archives, parquet file, and prior history were removed after this… See the full description on the dataset page: https://huggingface.co/datasets/webshart/terminusresearch-photo-aesthetics.qwen-image-2.1-generated-images
Synthetic image training set generated with Qwen Image 2.1
Built with Qwen. 10,000 generated images with their prompts, seeds, dimensions and generation metadata. This reusable dataset was prepared for assistant-LoRA training, so subsequent runs can train from cached images rather than regenerate a teacher target at every update.
Images were generated by the unmodified Qwen Image 2.1 base model, with 40 inference steps, CFG 1, BF16 and full-frame VAE decoding. No assistant or… See the full description on the dataset page: https://huggingface.co/datasets/webshart/qwen-image-2.1-generated-images.terminusresearch-photo-anatomy
Photo Anatomy — Webshart
This is the canonical, maintained location for the former terminusresearch/photo-anatomy dataset. The migration completed on August 19, 2026, with 15,425 image samples repackaged into 120 indexed Webshart shards. Every sample now includes both the original CogVLM caption and a newer GLM-5.3-Flash caption.
The legacy repository now contains only a relocation notice. Its original tar archives, parquet file, and prior history were removed after this copy… See the full description on the dataset page: https://huggingface.co/datasets/webshart/terminusresearch-photo-anatomy.cc12m-structured-captions
CC12M multi-caption Webshart indexes
Metadata-only indexes for the 10,994,853 available images in
laion/conceptual-captions-12m-webdataset, across 1,100 tar shards. The images
remain in that source dataset; this repository does not duplicate or rewrite
the image archives. Original member offsets, lengths, sidecar references, and
image geometry are preserved.
These per-shard Webshart indexes are based on
webshart/conceptual-captions-12m-webdataset-metadata.
The 1,100 JSON files… See the full description on the dataset page: https://huggingface.co/datasets/webshart/cc12m-structured-captions.OpenVid-1M-webshart-indices
OpenVid-1M webshart indices
Webshart indices for Dev-Jahn/OpenVid-1M-wds.
Each JSON file contains byte offsets and sample metadata for the correspondingly named uncompressed tar shard. Paths mirror the source dataset's train/ and val/ directories. The indices are generated remotely with webshart==0.4.6; progress.json records the pinned source revision and upload progress.
Use with webshart
Install webshart:
pip install -U webshart
Pass this repository as the… See the full description on the dataset page: https://huggingface.co/datasets/webshart/OpenVid-1M-webshart-indices.pseudo-camera-10k-structured
pseudo-camera-10k, native structured captions
The 9,997 images and structured captions from
terminusresearch/pseudo-camera-10k-structured-json, pinned to
revision 29b6406122ce01d82aa72148984e3c7ac7ae405e, repackaged into 19 indexed webshart tar shards.
The original photographs come from
bghira/pseudo-camera-10k.
License metadata is inherited from the source dataset.
Image bytes are unchanged. Each source .txt caption is parsed into a native
JSON object and embedded under… See the full description on the dataset page: https://huggingface.co/datasets/webshart/pseudo-camera-10k-structured.terminusresearch-photo-typography
Photo Typography — Webshart
This is the canonical, maintained location for the former terminusresearch/photo-typography dataset. The migration completed on August 19, 2026, with 13,167 captioned image samples repackaged into 209 indexed Webshart shards.
The legacy repository now contains only a relocation notice. Its original tar archives, parquet file, and prior history were removed after this copy was validated, avoiding a duplicate of roughly 222 GB. Existing users should… See the full description on the dataset page: https://huggingface.co/datasets/webshart/terminusresearch-photo-typography.e621-2024-webp-4Mpixel-webshart-indices
e621-2024-webp-4Mpixel webshart indices (shards 0-48)
Per-shard webshart index files for
NebulaeWis/e621-2024-webp-4Mpixel,
covering tar shards original/data-0000.tar through original/data-0048.tar
(193,966 samples). Each entry carries the tar member byte offset/length, image
geometry (width, height, aspect), and captions merged from
RareConcepts/e621-2024-metadata-4M-recap-qwen2.5-vl-7B
(qwen2.5-VL-7B recaptions).
Coverage is deliberately partial (first 49 of 1000 shards, built… See the full description on the dataset page: https://huggingface.co/datasets/webshart/e621-2024-webp-4Mpixel-webshart-indices.