File size: 1,811 Bytes
cb1678d
7fa3666
 
cb1678d
7fa3666
 
 
 
 
cb1678d
7fa3666
 
1ad7b8e
7fa3666
 
 
 
b08b37f
9e28ee3
b08b37f
33895f0
7fa3666
 
 
 
 
 
474d398
7fa3666
 
 
c7f2b16
 
 
 
7fa3666
 
4208f1e
7fa3666
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
---
language:
- en
license: creativeml-openrail-m
thumbnail: "https://huggingface.co/Avrik/abstract-anim-spritesheets/resolve/main/AnimationGrid.gif"
tags:
- stable-diffusion
- text-to-image
- image-to-image
---
**Abstract Animation Sprite Sheets**

An experimental Dreambooth model trained on individual frames of looping 3D animations that were then laid out on a 4x4 grid. Generates sprite sheets that can create very interesting abstract animations.

Use the token **AbstrAnm spritesheet**. Size must be set at 512x512 or your outputs may not work properly.

Example prompt:
**AbstrAnm spritesheet, animation of a red glowing orb in the sky, highly detailed, fog, atmosphere, glow, sprites, animated, abstract**

**Negative prompt: high contrast, text, overlay**

Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 8

Feel free to experiment with other types of prompts and/or model merges.

![Sample Generations](https://huggingface.co/Avrik/abstract-anim-spritesheets/resolve/main/AnimationGrid.gif)

You can also upscale it 4x to produce 512x512 animations. Used SD Upscale from AUTOMATIC1111's web UI to add more sharpness and detail.

![Upscaled](https://huggingface.co/Avrik/abstract-anim-spritesheets/resolve/main/AnimationGridUpscale.gif)

Further testing also made me discover it's quite flexible and could even animate less abstract concepts. There's actually quite a lot you can do with this, though not nearly as coherent or powerful as a fully trained text2video model.

![New Animations](https://huggingface.co/Avrik/abstract-anim-spritesheets/resolve/main/natureanims.gif)

Some issues with this model.
- May not loop seamlessly
- Tends to be too noisy
- Sprites aren't usually perfect squares
- Small size and short animation (could experiment with training on larger resolutions in the future)