Avrik's picture
Update README.md
c7f2b16
|
raw
history blame
1.81 kB
---
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)