---
inference: true
language:
- en
tags:
  - stable-diffusion
  - text-to-image
pipeline_tag: text-to-image
---

This is a Stable Diffusion model and collection of embeddings which were trained off of Bluey characters, and can generate semi-consistent and accurate results. While nowhere near perfect, its a start, and has certainly been fun being able to work on.

# It is absolutely essential you *use both the checkpoint AND embeddings.* 

The way I trained it is a little unconventional, and you will struggle to get decent results without all the components.
The base understanding of character shapes and style was learned in the checkpoint.
The individual character colors, patterns, proportions, and extra refinement was applied through embeddings.

## Examples of generated results:
![ReferenceImage](RefImages/1.png)
![ReferenceImage](RefImages/2.png)
![ReferenceImage](RefImages/3.png)
(note that these images were cherry-picked, don't be shocked to see some cursed abominations)

# Example negative prompt:
```
weirdly colored eyes, weird eyelids, bad art, ((inaccurate)), (missing pupils), weird pupils, inaccurate eyes, weird eyes, bad proportions, more than two ears, (less than two ears), blank eyes, missing hands, weird, inaccurate pupils, colored eyes, (missing tail), ugly, weird scaling, deformed, disfigured, stretched, tiled, unofficial, unofficial artwork, poorly drawn, blurry, (pixelated), pink, purple, letterbox, border, image artifacts, compressed
```

# limitations:
- Everything was trained from 512x512 images, higher resolution generations will have issues
- It does not know what the characters look like from behind, and only generate images of them looking forward, left, or right.
- It struggles to get more dynamic poses, it does good at standing, mediocre with sitting, but laying down and others it struggles.
- Often gets confused with eyes and eyelids
- Bingo embedding has less training, and often confuses her for Bluey.
- Difficult getting multiple different characters in a single image consistently without oddities.

# **(Disclaimer)**

*I am not affiliated with Ludo and do not own the rights to Bluey or its content, this was trained off of images from Bluey of which I do not own, and produces images similar to Bluey.
This was made purely from an educational, research, and general fun perspective. I have the highest level of respect for Ludo and their intellectual property, and do not in any way take credit for their work.*