|
--- |
|
license: creativeml-openrail-m |
|
tags: |
|
- text-to-image |
|
- stable-diffusion |
|
--- |
|
### Wriggly's-Extra-Gum-artstyle-waifudiffusion Dreambooth model trained by Kagerage with [TheLastBen's fast-DreamBooth](https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast-DreamBooth.ipynb) notebook |
|
|
|
|
|
Test the concept via A1111 Colab [fast-Colab-A1111](https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast_stable_diffusion_AUTOMATIC1111.ipynb) |
|
Or you can run your new concept via `diffusers` [Colab Notebook for Inference](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/sd_dreambooth_inference.ipynb) |
|
|
|
As the name suggests, this model was trained on that weird hand-drawn looking artwork that you'd find on the inside of a pack of Wriggly's Extra gum, ontop of Waifu Diffusion 1.3. |
|
|
|
This model is pretty strange to use, due to how homogenous the training dataset was. Despite that, it is quite capable of producing really stylized images. For best results, try moving the trigger word "gvgtgm" to various parts of the prompt, sometimes adding "anime illustration" or "anime drawing" just after the trigger word, maybe mention either "black and white" or no color. Negative prompts don't seem to do much in terms of locking in the style, though they still work for things like retaining a normal limb count. My best results came using prompts like "one piece style gvgtgm,", and "ojamajo doremi style gvgtgm,", though feel free to experiment. |
|
|
|
Recommended sampler is anything non-ancestral, as to avoid softness. CFG 7 seems to work fine. Looks like it generates decent results without the Waifu Diffusion VAE, not sure how it acts with it enabled. |
|
|
|
Sample pictures of this concept: |
|
|
|
|
|
|
|
 |
|
 |
|
 |
|
 |
|
 |
