|
--- |
|
language: |
|
- en |
|
tags: |
|
- stable-diffusion |
|
- text-to-image |
|
license: bigscience-bloom-rail-1.0 |
|
inference: false |
|
|
|
--- |
|
|
|
# Doodad |
|
|
|
*<p style="color:green"> <b> This is the 1st part of a 2 (3?) part project meant to be used with <a href="https://huggingface.co/doohickey">Doohickey</a> </b> </p>* |
|
|
|
The model was trained with [Dreambooth-Stable](https://www.reddit.com/r/StableDiffusion/comments/xphaiw/dreambooth_stable_diffusion_training_in_just_125/). It mixes especially well with the <midj-strong> style included in Doohickey. It's a finetuned version of the Trinart-Waifu-diffusion-50-50 included in this organization's models and was trained on 48 images from the author's ([crumb](https://huggingface.co/crumb)'s) Pinterest feed. |
|
|
|
| Parameter | Value | |
|
|----------------------|------------| |
|
| resolution | 512 | |
|
| train_batch_size | 1 | |
|
| gradient_accumulation_steps | 2 | |
|
| learning_rate | 5e-6 | |
|
| num_class_images | 120 | |
|
| max_train_steps | 1200 | |
|
|
|
Example outputs: |
|
|
|
"portal to another dimension" |
|
 |
|
|
|
"a portrait of a dying god" |
|
 |
|
 |
|
|
|
"photograph" |
|
 |
