Fer14/colored-dragons

Colored Dragons

About 7,400 synthetic 1024×1024 PNG images of Dyno, the cartoon dragon from DYNO, in many colours and variations.

This is the dataset used to train the DYNO VAE. In the game, the VAE decoder turns a 1024-dim latent vector into a unique dragon you can hatch, breed, battle and share by QR code.

The original Dyno

All images are variations of this base dragon:

Showcase

Red	Orange	Gold	Black
Teal	Blue	Purple	Magenta

Structure

Every image is in a single folder, with no labels or metadata:

images/
  7a6fc4e58020ee37506846525fe97e05.png   # the original Dyno
  *.png        # ~7,400 colour/pose/feature variations

File names are the MD5 hash of each image's original file name.

Usage

from datasets import load_dataset

ds = load_dataset("Fer14/colored-dragons", split="train")
ds[0]["image"]

How it was made

The images were generated from original.png with image-editing models: hand-made colour variants, plus large batches from FLUX.1 Kontext prompted for different colours, horns, wings, sizes and other features. Every image keeps the same pose and style as the original Dyno.

Links

• Game and training code: github.com/Fer14/dyno