---
license: gpl-3.0
datasets:
- pawkanarek/spraix_1024_9frames
library_name: diffusers
pipeline_tag: text-to-image
tags:
- art
---
### Why
My intention was to creat model that could generate animated sprites.

### How 
Trained for ~48 hours in total 25 epochs, on Google TPU v3-8 with this my custom script [train_text_to_image_flax_sdxl](https://github.com/PawKanarek/spraix/blob/48d8c209a359622e6db56e6d555667ac466dc952/train_text_to_image_flax_sdxl.py) on my custom dataset [spraix_1024_9frames](https://huggingface.co/datasets/pawkanarek/spraix_1024_9frames). 

This is my second attempt to create such model. This time i changed dataset to always consist 9 frames, but it doesn't help much.

### Appendix

This is demonstration only. This is my first time where i wrote training script of such complicated model in flax framework. Scirpt is probably full of bugs. Dataset that i prepared is also far from perfect. But I'm happy that i can train, save, and load my custom SDXL model.

###   How to use 
Intented to use with FLAX

Model will generate ugly sprite animations, out of shape, deformed, pixelated and uneven.
Prompt ideas
```
"Pixel-art animation of a blue water droplet with legs, that: is swiniging axe, facing: East",
"Pixel-art animation of a Tree, that: is Idle, facing: South",
"Pixel-art animation of a Dinosaur with a backpack, that: is jumping, facing: North",
"Pixel-art animation of a Fire demon with axe, that: is running, facing West",
```

Samples of the output of this model can be seen here: