All the intermediate checkpoints can be found in https://huggingface.co/alea31415/tenten-characters
The base model is ACertainty
Trained at clip skip 1.
For LoHa you may need to adjust weight according to the base model that you are using, and switch to an earlier checkpoint (tenten_loha_step-20000.safetensors or tenten_loha_step-30000.safetensors) may also help.
If you find the anime style too strong you can try to put aniscreen in the negative but with a small weight.
On the other hand putting aniscreen in the positive recovers the full anime style.
Example Images
LoHa: 2,3,9,10,12,13,14,15,16,18
Checkpoint: 1,4,5,6,7,8,11,17,19,20
--
I don't need support or credit, but I would be glad to know that you are using the models I trained and find it useful.
Moreover, I would like to advocate for more franchise models.
You can take a look at my workflow https://github.com/cyber-meow/anime_screenshot_pipeline if you are interested.
I just want to spread the fact that there is no reason to encode a single concept in each lora.
Description
FAQ
Comments (6)
可以問一下大佬怎麼從Lora變成LYCORIS了? 這兩個的差別是甚麼?
https://github.com/KohakuBlueleaf/LyCORIS
LyCORIS 是一個 sd fine-tune 的 project
使用的時候安裝 civitai 標明的 extension 就好
目前有
- LoCon, 在原版的 lora 上面補完 residual block, 這個後來 kohya 也加了
- LoHa (這裡使用的) 差別在於 fine-tune 的時候 weight difference 拆解的方式從單純 low rank decomposition 變成 hadamard product of two low rank matrices
另外跟 kohya 的比 LyCORIS 目前 conv block 的拆解方式 default 換成了 cp decomposition
最後 extract 支援 sparse bias
大概是這樣 接下來要加什麼 feature 還需要思考
This looks very very very good.
However, I have tried many settings and could only produce poor quality Anisphia.
Maybe I am missing some information on this page.
For example, what VAE do you use?
The anything VAE. Can you give more details? Are you using full checkpoint or loha? What do you use as prompt and what is your base model if you are using loha? Do you effectively use clip skip 1?
You should be able to generate good images with all the models I provide here though, example for loha
https://i.imgur.com/fVWZ9DR.jpg
@alea31415
The anything VAE!
I had previously downloaded it on Hugging but had not been able to activate it.
I downloaded Anything-V3.0.vae.pt again on civitai and was able to activate it.
As a result, I could create the piece I want. It is very wonderful.
I used tenten0318-ep00-gs16582.safetensors.
Your model is my favorite of the models available on civitai.
And all your models look great.
Your additional information is also very helpful. I will pick up a combination that looks good and try it out.
Thank you.
Looks great, can't wait to try. Thanks for the girlpack!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.