Skip to content
Bluzen Labs
All projects
DataActive

Aurora Corpus

The dataset everything else is trained on

An 18M-image illustration corpus with tags, captions and quality bands, plus the whole filtering pipeline that produced it.

Category
Data
Status
Active
Language
Python
Licence
CC-BY-4.0
Stars
980
Since
2026

A dataset without its pipeline is an artefact you cannot audit. Aurora publishes both: the parquet shards, and the filter chain that decided what got in.

The chain

  1. Crawl and archive with source attribution retained.
  2. Perceptual dedup across the whole corpus.
  3. Tagging, captioning and aesthetic scoring.
  4. Quality banding, published as a column rather than applied as a deletion.

The last point matters. Filters are opinions. Shipping the score instead of the deletion lets someone else disagree with ours.

Status

Active. Growing roughly a million rows a month.

Writing

Data6 Aug 2026 · 6 min

Filters are opinions, so ship the score

On building Aurora: why every quality filter in the pipeline emits a column instead of deleting a row, and what that costs in storage.

Ashish Kumar, Yuki HoshinoRead

Related projects

BluzenTagger

Multi-label tagging for illustration

Active

A tagger over a 12k-tag vocabulary that holds up on the long tail, where general-purpose vision models quietly collapse into the fifty most common labels.

#PyTorch#ONNX#Vision#Tagging
Language
Python
Stars
1.4k
Licence
Apache-2.0
Category
Research

MochiCore

Adapter training that fits on one GPU

Stable

The trainer the rest of the lab runs on. Adapter families, quantised paths and a config format you can read out loud.

#PyTorch#LoRA#Training
Language
Python
Stars
3.6k
Licence
MIT
Category
Training

PrismStream

Flow-matching generation with region control

Active

A text-to-image stack built around region conditioning, so a prompt can say where something belongs rather than only what it is.

#Diffusion#PyTorch#Vision
Language
Python
Stars
2.1k
Licence
Apache-2.0
Category
Research