You don't even need to write any positive prompts to create beautiful girls with this model, which was created with many photos and many stages. I've included a few comparison charts below to show how it differs from some other realistic models.
1) No positive prompts were used in this comparison.
2) In this comparison, only the word "2girls" was used as a positive prompt.
3) In this comparison, only the word "3girls" was used as a positive prompt.
I use ADetailer to beautify faces.
I also find it useful to use negative embeddings such as instarealneg, SkinPerfection, np_simple_negatives.
I use Color101VAE but you can try different things.
Setting Clip Skip to 1 gives more successful results.
Good results can be obtained at resolutions such as 384X680, 416X624, 440X584, 456X568, 512X768 and 640X960. Or vice versa for horizontal images... But the success rate may decrease slightly for horizontal images.
I definitely recommend using Hires fix, but set the denoising strength to a value around 0.1. When you increase the value, facial expressions change a lot. Of course, if you use ADetailer, you can increase this value as you wish. ADetailer already adjusts the faces itself.
Description
FAQ
Comments (14)
Definitely trying this! i love how it's captured the british|Irish look of women, the hair extensions the makeup & tan lol, it's actually a unique model because of this the site has so much anime and american looking women gens so A++ for those displayed images
what is 'instarealneg'?
A negative embedding that was published on civitai a while ago... I think they deleted it later. You can download it from:https://drive.google.com/file/d/10HKcSpg-iV9VYCfVUilopmU-tdEIKLJy/view?usp=drive_link
@53rt5355iz appreciate that
When trying to use this model in EasyDiffusion, I get "Error: Could not load the stable-diffusion model! Reason: 'Attention' object has no attribute 'to_to_k'"
I've never used EasyDiffusion. I don't know the working principle. Maybe this model is not suitable for that platform. It works fine in Automatic1111 or ComfyUI.
This can be fixed by running the following script from the Easy Diffusion Developer Console:
https://github.com/huggingface/diffusers/issues/4790
1. Load the developer console
2. python [enter]
3. Remove all empty lines in the script at the bottom of the above link, then paste the text into the console.
4. Insert your file paths:
original_file = '/path/to/beautyfoolGirls_v10.safetensors'
fixed_file = '/path/to/beautyfoolGirls_v10_fixed.safetensors'
5. Run the fix:
fix_diffusers_model_conversion(load_path=original_file, save_path=fixed_file)
@iwasrobert I was able to run this directly in SwarmUI . Back when I was using only EasyDiffusion, I asked this question and I am happy that it got an answer. I have not tried to run the conversion with EasyDiffusion, since I feel the results I get from this checkpoint is not as good as I can get with other checkpoints, e.g. realisticVision v5.1 https://civitai.com/models/4201?modelVersionId=130072 - I will have to try to tweak some embedding or add some loras to get the UK Girl feel in the photos, but I think I will be able to do it. Many thanks for pointing out the issue on github with the code needed to convert this in a way that makes it possible to run with EasyDiffusion. If I do run the conversion at some point, I will let you all know how it went.
@iwasrobert thank you for pointing out this solution, but I am having issues getting it to work. I get indent errors in the console. I'm also not positive I'm inserting the paths in the correct location in the code as I don't see the actual phrases "original_file" or "fixed_file"
Let's try this.
@iwasrobert thanks for offering help. Later in the evening yesterday I had luck using the colab version of the code posted on the link you sent. File paths were cleared to me and it was able to help me fix dependencies I didn't understand.
Where can I find "np_simple_negatives_v2-neg"?
I think they deleted it, but you can find a different version here: https://civitai.com/models/87243?modelVersionId=287973
@53rt5355iz Oh~thanks