---
license: mit
task_categories:
  - text-to-image
  - image-to-text
language:
  - en
  - tr
size_categories:
  - 10K<n<100K
tags:
  - aurora
  - diffusion
  - text-to-image
  - image-generation
  - prompt-gallery
  - synthetic
  - turkish
  - tr
  - 64x64
pretty_name: Aurora Mini Diffusion — Curated
---

# Aurora Mini Diffusion - Curated

Quality-filtered subset of the generated images: CLIP ViT-B/32 text-image alignment score >= 0.33, plus aesthetic heuristics (saturation, color variance, brightness) and Laplacian blur detection.

## Statistics
| Field | Value |
|-------|-------|
| Images | 17,298 |
| Unique prompts | 1,610 |
| Resolution | 64x64 |
| Format | JPEG (quality 88) |
| Source | AuroraIMG-6M (AuroraAI-Research) |

## Schema
| Column | Type | Description |
|--------|------|-------------|
| `image` | Image (JPEG) | 64x64 generated image |
| `caption_en` | string | English prompt used for generation |
| `caption_tr` | string | Turkish translation of the prompt |
| `seed` | int | RNG seed used for generation |
| `prompt_idx` | int | index of the prompt in the generation prompt list |

## Generation
Images generated with [AuroraIMG-6M](https://huggingface.co/AuroraAI-Research/AuroraIMG-6M)
(6.8M params, flow matching, Euler, 8 steps, CFG 6.0) from 10,000 diverse English
prompts x 20 random seeds.

## License
MIT