New additions -
Ageslider for PONY
Brightness for SDXL and PONY
NoseSize for SDXL and PONY
Sliders uploaded last year were shared by Rohit Gandikota. All credits goes to him, thanks for sharing these new sliders. All I did was to convert to safetensor files. The original pt files were shared here
Description
FAQ
Comments (41)
thank you for converting them into safetensors, I believe there were no ill intentions with the other files, but for safety and considering how simple it is nowadays to have either pickle or safetensors (with plenty of converters) the later are always preferable
I get that. normally I would have converted them earlier when there are multiple users asking for them, given it's an easy request, uploading these files to civitai took much more time than converting them. I simply didn't want rude people to think they always get what they want by threating creators instead of asking nicely. But there are more nice people than assholes, I probably shouldn't do it out of spite.
@NextMeal i understand that, i see so many "why is this sdxl, make it 1.5" or vice versa, a full training of any model be it for one or other takes time, when getting models nowadays tho, safetensors are the way and from experience (I used to have a too weak pc so would need to actually use pickles on 1.5 for everything) the conversion is not too resource intensive, i would not ask for sth like a full retraining or other sd version (that is basically the same) of a model
tried to do a large range comparison between 12 of em and upload but I'm guessing that due to the age slider (since it shows age and a low number) it doesn't allow the upload due to ToS violation -.-
@NextMeal Hey boss your hair slider is actually age. can you upload the hair slider?
Wow! Hey, thanks so much! That was unbelievably kind of you. Shows the spirit of the season! I'm sorry the other person, whoever that was, was a bit argumentative and demanding. I thought if I was lucky, maybe you'd share a link on how one coverts things, which I admit, would be entirely new territory for me - but this just blew me away and is a welcome reminder that there are really kind people out there. I applaud you, kind sir/madame. Well played.
Now I can play with these .safetensors in Fooocus now. Thanks for doing this, you're great!
Given their size, are sliders basically amplifying concepts that the model already knows?
yes, indeed. Emphasis/Reduce concepts that models already know OR concepts that are hard to prompt. could be done for both.
I have tried some of your loras.
Age, Weight, Long Hair, Muscular does not work for me on Fooocus,. Nothing happens.
Smiling works, Surprised Look works, Festive Works
I think maybe you didn't use enough weight. I didn't write down a suggested weight because each version has different preferable range. It can be -2 ~+2 or -7 ~ +7. If you use weight = 1, it barely does anything. I didn't make these, but they work well in my testing, theoritically they should work better than sliders trained with older methods.
Thank you. That worked!
You missed a couple, but thanks!
Thanks, I know there are still a couple, but I don’t find they produce expected results even in base model. (Maybe my prompts were too simple? Or they need special settings) I couldn’t generate usable sample images for them, so I left them out.
Thank you very much for taking the time to find and add these for us, its really appreciated and will be really helpful in my creative project! And happy new year :)
the sliders all turn men into women....
how is that even possible? The models weren’t trained on females. are you sure it's not a checkpoint or prompting issue?
@NextMeal they weren't trained on females? Strange. I'm using male-focused models and they're getting makeup and braids. I'll try again.
You used males in training?
Hi NextMeal, I've tried to install theses in DrawThings app, but for some reason they are not recognized as LoRAs, and the .checkpoint does not install...
Any thoughts on why that would happen?
Same observation by me.
@DrDoItYourSelf Same here
Can you point to documentation for how you converted the .pt to safetensors?
import torch import os from natsort import natsorted print(torch.version.cuda) print(torch.cuda.is_available()) from safetensors.torch import save_file root_folder = '.' for filename in os.listdir(root_folder): # Check if the file is a .pt file if filename.endswith('.pt'): # Load the .pt file file_path = os.path.join(root_folder, filename) model = torch.load(file_path) # Convert the filename to .safetensors new_filename = filename.replace('.pt', '.safetensors') new_file_path = os.path.join(root_folder, new_filename) # Save the model in SafeTensors format save_file(model, new_file_path) print(f'Converted {filename} to {new_filename}')
I had a simple script, sorry about the formating.
@NextMeal Could you share the script?
@Beethy The script was posted, the comment is literal. However, it has removed line breaks, so it's a bit strange. It appears to be Python.
Thanks :)
Thank you very much for your work. Small observation (in my setup at least) I'm finding that negative values tend to wash out the picture, and positive values tend to progressively darken it - which seems a great pity, 'cos for the hands themselves it's a big step in the right direction.
I took some time to make examples.
Some of them are very easy to use, but I didn't take time to understand others and make examples. When I experiment with them I'll be sure to come back and post those examples too.
Thanks again!
I hope you make some PONY slider too. Eye size slider would be great, but it doesn't work outside of SDXL :/
I feel exactly the same way!
Eye size slider is so Great, Please make one for PONY as well.
The age and hair length sliders have the same hash sum, I think there's the hair LORA instead of the age one uploaded here.
Do these also work if you're trying to get a young age such as with the children's storybook checkpoint?
It doesn't work well on the pony model. Only with a weight of 5+ does it begin to sharply turn into a decrepit old woman. Finding the middle is extremely difficult.
That's not surprising at all, to be honest.
The original SDXL model is very bad at NSFW, while Pony is very flexible at NSFW content. To achieve that, Pony's creators "overtrained" Pony, which turned it into a model considerably different than SDXL. Because of that, resources (checkpoints, loras, etc) trained for SDXL will work poorly (or not work at all) with Pony, and vice-versa.
That's why the site classifies resources meant for Pony and resources meant for SDXL, just like it classifies resources meant for SDXL and those meant for 1.5.
can someone explain how to use it?
age has a two LoRA's which one is a duplicate for long_hair. Also cartoon and pixar are identical.
Now age has 2 identical long_hair loras, no age lora.
i need tips is there set trigger words, sick of kid faces set weight on lora at :2 still kids writen age old
u mean like the elderly people still look too youngish? probably an inherent problem with AI. ever seen those "this is the visually averaged face of X country!" and they're all young and beautiful? AI has Amalgam Bias, there's a reason some people make "ugly/average people" LoRas and they're all still pretty decent looking (depending on taste). the only 'ugly' i've seen in models are horror themed, and even then, there's not some vast IRL variety of features and types
I have tried age and chubby/weight loras. In comfy none seemed to work. Tested with regular and inpainting checkpoints. Since there're some positive reviews I'm wondering if i'm doing something wrong or ppl use it in different tools?