|
--- |
|
datasets: |
|
- KBlueLeaf/danbooru2023-sqlite |
|
language: |
|
- en |
|
library_name: diffusers |
|
--- |
|
|
|
# Kohaku XL Gamma |
|
|
|
A SDXL anime base model aims to create unique artworks. |
|
<div style="display:flex; flex-direction: row; width: 100%;"> |
|
<div><img style="display:block;" src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/afc1dc39-3a1c-4db9-9942-0b2c150124ce/original=true/00946.jpeg"/></div> |
|
<div><img style="display:block;" src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/20751b43-3d91-4802-b5cb-ce3857d1027a/original=true/00937.jpeg"/></div> |
|
<div><img style="display:block;" src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/fade32c7-7cb8-4a69-84f3-d3da0f4b319d/original=true/00943.jpeg"/></div> |
|
</div> |
|
|
|
--- |
|
|
|
## Introduction |
|
|
|
This model can be seen as a derivative of Animagine XL 3.0 project.<br> |
|
Basically I'm collaborating with Linaqruf for making better Anime base model (and it is obvious that we have different goal/target)<br> |
|
We share our models and technique to improve our models' quality.<br> |
|
And that is also how this model been created.<br> |
|
|
|
|
|
### Base7 |
|
|
|
Kohaku-XL base7 is resumed from beta7 and use same dataset that beta series have used. But this time I use my own metadata system to create captions. (Can be taken as advanced version of what linaqruf used, will open source it soon) |
|
|
|
Trainin details:<br> |
|
LR: 8e-6/2e-6<br> |
|
Scheduler: constant with warmup<br> |
|
Batch size: 128 (batch size 4 * grad acc 16 * gpu count 2) <br> |
|
|
|
|
|
### Gamma rev1 |
|
|
|
Kohaku-XL Gamma rev1 is a merged model which combine the learned diff from anxl3 and kohaku xl base 7. With this forumla: |
|
|
|
gamma rev1 = beta7 + 0.8 * (anxl3 - anxl2) + 0.5 * (base7 - beta7) |
|
|
|
--- |
|
|
|
## Usage |
|
#### Parameter |
|
|
|
This model is trained under 768x1024 to 1024x1024 ARB. It is recommended to use pixel count within 786432 ~ 1310720.<br> |
|
Recommended CFG scale is 4~7.<br> |
|
Sampler should not matter.<br> |
|
|
|
#### Tagging |
|
This model use my own system for quality tags or something like that. <br> |
|
So although this model combine the diff weight from anxl3, I will still recommend user to use mine (or both) tagging system.<br> |
|
|
|
The format of prompt is as same as anxl3. (You can check the sample images I post) |
|
|
|
Rating tags: |
|
* General: safe |
|
* Sensitive: sensitive |
|
* Questionable: nsfw |
|
* Explicit: explicit, nsfw |
|
|
|
Quality tags (Better to worse): |
|
* Masterpiece |
|
* best quality |
|
* great quality |
|
* good quality |
|
* normal quality |
|
* low quality |
|
* worst quality |
|
|
|
|
|
Year tags (New to Old): |
|
* newest |
|
* recent |
|
* mid |
|
* early |
|
* old |
|
|
|
|
|
You may meet some subtle mosaic-like artifact, that may be caused by high-lr or bad resizing/image encoding.<br> |
|
I will try to fix it in next version. For now, try to use R-ESRGAN anime6b or SCUNet models for fixing it.<br> |
|
|
|
--- |
|
|
|
## Future plan |
|
|
|
Since my dataset have some resize/webp artifacts that will harm the models. I will recreate my dataset based on my new system (and opensource it once I done it). |
|
|
|
The next plan is to train model on larger (3M-6M) dataset with better configuration (which will require A100s and I plan to spend about 2000-10000 USD on it, if you like my works, consider to sponsor me via buy-me-a-coffee or some BTC-sutff) |
|
|
