|
--- |
|
license: other |
|
license_name: faipl-1.0-sd |
|
license_link: https://freedevproject.org/faipl-1.0-sd/ |
|
pipeline_tag: text-to-image |
|
base_model: |
|
- OnomaAIResearch/Illustrious-xl-early-release-v0 |
|
tags: |
|
- stable-diffusion |
|
- stable-diffusion-xl |
|
--- |
|
|
|
# Millennium-IL |
|
|
|
![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/630e2d981ef92d4e37a1694e/THPS9LiA8cHY0N9SRgnTc.jpeg) |
|
|
|
Modified Illustrious-XL-v0.1 with Blue Archive style |
|
|
|
Do not expect quality! |
|
|
|
## Prompt Guidelines |
|
Almost same as the base model |
|
|
|
## Recommended Prompt |
|
None(Works good without `masterpiece, best quality`) |
|
|
|
## Recommended Negative Prompt |
|
`worst quality, low quality, bad quality, lowres, jpeg artifacts, unfinished, monochrome` |
|
|
|
## Recommended Settings |
|
Steps: 14-28 |
|
|
|
Sampler: DPM++ 2M(dpmpp_2m) |
|
|
|
Scheduler: Simple or Karras |
|
|
|
Guidance Scale: 4-9 |
|
|
|
|
|
## Training information |
|
Finetuned Illustrious-XL-v0.1 by repeating the training and merging a DoRA 8 times with sd-scripts. |
|
|
|
- Network module: lycoris_kohya(algo=lora, dora_wd=True) |
|
- Resolution: 1024(Bucketing enabled, min 512, max 2048) |
|
- Optimizer: Lion |
|
- Train U-Net only: Yes |
|
- LR Scheduler: cosine with restart(warmup steps=80-150, repeat=4-6) |
|
- Learning Rate: various(min=1.5e-05, max=7e-05) |
|
- Noise Offset: 0.04 |
|
- Immiscible Noise: 2048 |
|
- Batch size: 1 |
|
- Gradient Accumulation steps: 1 or 2 |
|
- Dim/Alpha: 16/4 |
|
- Conv Dim/Alpha: 1/0.1 |
|
|
|
## Dataset information |
|
Dataset size: 338 |
|
|
|
## Training scripts: |
|
[sd-scripts](https://github.com/kohya-ss/sd-scripts) |
|
|
|
## Notice |
|
This model is licensed under [Fair AI Public License 1.0-SD](https://freedevproject.org/faipl-1.0-sd/) |
|
|
|
If you make modify this model, you must share both your changes and the original license. |
|
|
|
You are prohibited from monetizing any close-sourced fine-tuned / merged model, which disallows the public from accessing the model's source code / weights and its usages. |