Support☕ https://ko-fi.com/sfa837348
hugging face https://huggingface.co/gsdf/CounterfeitXL-V2.0
Negative Embedding https://civarchive.com/models/118418/negativexl
Official hosting for online AI image generator.
A small update.
V2.5 is a revised version of V2.0.
This version has been retrained for style tags.
Tips
By inserting 'flat color, anime style' into negative prompts, you can achieve results with higher detail.
CounterfeitXL-V2.0 has been primarily trained on the latest illustrations but encounters a recurring issue with the appearance of halos. In such cases, please include 'halo' in negative prompts.
I have incorporated style tags into some of the training images. By describing these tags in the prompt, various effects can be achieved. Additionally, inserting style tags into negative prompts or combining them allows for the generation of images that align with your imagined.
Style Tags
masterpiece: Beautifully and intricately drawn illustrations.
cinematic style: Stunning illustrations with detailed backgrounds and lighting.
aesthetic style: Illustrations featuring beautiful color schemes.
cute style: Cute.
flat color: Illustrations with strong lines and a minimal color palette.
anime style: Anime.
comic style: Illustrations characterized by bold color choices and abstract representations.
Description
FAQ
Comments (38)
great！
I hope to have more information about this model
Awesome!
hope to see counterfeit β soon
Should we use the SDXL refiner or not for this model?
Sadly it will not work proper u can read my summary on that
The results with and without refiner are not close to the sd1.5 outputs. The outputs counterxl and refiner 1.0 only I havent postes bc output is really out of standard. Counterxl, refiner 1.0 and upscale with counter1.5 are also not good. I think its necessary to train refiner and base model otherwise trained 1.5 are superior.
My summary: I have tested the trained xl models published recently on civi and the native xl base model with several different workflows (With, without refiner, refiner sd1.5, base and refiner xl with upsclae sd.1.5. Conclusion: xl base and refiner xl are useable for some outstanding outputs. But the trained sd.1.5 models are superior to their xl variants and my opinion switching to trained xl models makes only sence if there is also an appropriate trained refiner available because the standard 1.0 refiner doesnt fit to the trained xl models. Just compare ure results gen by SD1.5 models and the corresponding xl model and u will know what I am talking about.
Perhaps loras require a refiner lora along with a standard concept lora.
点赞！正缺SDXL的动漫模型呢！
this embedding how to use?
when i put negativeXL_A on embeddings file just error
at a1111
This is not an embedding. This is a checkpoint model. You're getting this one confused with https://civitai.com/models/118418/negativexl
@theunlikely i using huggingface embeddings
show me this error
*** Error loading embedding negativeXL_A.safetensors
Traceback (most recent call last):
File "G:\ai3\stable-diffusion-webui\modules\textual_inversion\textual_inversion.py", line 227, in load_from_dir
self.load_from_file(fullfn, fn)
File "G:\ai3\stable-diffusion-webui\modules\textual_inversion\textual_inversion.py", line 192, in load_from_file
assert len(data.keys()) == 1, 'embedding file has multiple terms in it'
AssertionError: embedding file has multiple terms in it
@zens here is my fix:
1. git stash
2. git checkout 6f0abbb
do this on your root sd folder
@shiowoneko tq,is work for me
@zens indonesian eh?
Please help to sync and upload to the tensor website
After testing by members of the SD_CN QQ Group and my speculation: this model is either seriously overfitting or has a problem with the training method. The resulting images do not follow the prompt words at all.
Are you sure you aren't seeing the general problem with SDXL and how it'll ignore most the concepts of your prompt?
Please provide evidence on whether it is a valid test.
Not sure. See this post:
https://www.reddit.com/r/StableDiffusion/comments/15f6sd9/superheroes_villains_and_dragonball/
(Not mine btw)
@xexkenxex i mean he is prety credible if i say so... he made anyv3 and said that any v3 is a bad model in terms of promting due to overfitting and putting random stuff that it is not meant to........... and then he made anyv5 wich is imo the most balanced anime model out there
The reason I like this model is that it's overfitting :) It's one of the few that you can minimally noise for image to image and still get a strong anime style, especially for faces - eg you can follow the input image very closely and still get anime out. You can try 10 other anime models and not even come close to this one on that front.
Struggles quite a bit with more complicated concepts which is expected for a finetune of a non anime base. Despite that it's the best XL anime finetune so far and I'm excited to see what will be possible once wdxl comes out and can be used as a base.
Here we go, let the anime for XL begin
感觉画风不稳定，很大的可能出写实的图片。测试了几个关键词，只有作者给的几个示例效果好
I wonder which VAE do you use?Thanks
Would assume it's embedded
There is a CFXL VAE in the huggingface repo
There is a CFXL VAE in the huggingface repo
Hey, Can you please tell me if the anime in XL will be better than the just Stable Diffusion? I just don't know if I should try it
I tried XL, but I wasn't able to consistently create good anime images with any of the models. I tried training loras, but none of them turned out well like my 1.5 loras. I think it's because of the lack of a good anime model, which will come eventually.
how to past generation data on comfyui?
drop a png created by comfy into the workspace after hitting clear
Where should I place the embeddings file? Using SD v1.5.1, if I put it in embeddings folder, SD just throws AssertionError.
you need to use dev branch
1.5.1 is not yet supported
This is for SDXL 1.0, the newest version of Stable Diffusion