Styles are trained on synthetic data and do not copy the styles of artists!
Photo 2 - is trained on real data (Dataset from Unsplash)
[M] Merged LoRAs
Recommended LoRA Strength (Weight): 0.8-1
Preview generated without using ADetailer & Hires. fix
Description
Repeats: 10
Epoch: 4
Image Count: 72
Network Rank (Dimension): 32
Network Alpha: 16
Batch size: 1
Clip skip: 1
Training steps: 2880
Resolution: 1024x1024
Train Resolution: 1024x1024
LR Scheduler: cosine_with_restarts
Optimizer: AdamW8bit
Learning rate: 0,0005
Text Encoder learning rate: 0,00005
Unet learning rate: 0,0005
Model: ponyDiffusionV6XL_v6
Captioning: WD14 Captioning (SmilingWolf/wd-v1-4-moat-tagger-v2)
Trigger Words: painterly, faux traditional media, realistic,
Recommended LoRA Strength (Weight): 0.7-1
The same dataset as that of Faux Oil Painting, only less contrast and saturation
FAQ
Comments (50)
Amei
this is good
Excellent work
I would have to say that this is a must-have for elevating the creations in your repertoire.
Nice
good
out of curiosity, how do you get such consistent gens of the roughly the same image? Like do you make a good image, and then use control net or something to restyle it?
Take a look at the prompt generation details. Looks like they mostly keep the same details. Biggest takeaway though is probably consistent seeding.
good
Horny
I wonder if the "1girl, witch_hat, flower, etc" as a standard test prompt is related to "Lenna", the Playboy pinup image used as a standard test image in image processing research since the early '70s? Maybe a coincidence...
G.O.A.T.
I use twilight and picture have texture like noise and spot, but adetailer zone very smooth. I use on A1111 set CLIP Skip 2 doesn't help. I don't want to use on comfyUI. Anyone can help me?
CLIP SKIP does not matter in A1111/Forge for SDXL, the value can be anything and it will not change the situation
I use Forge, which is practically the same as A1111 and I have no problems using my LoRAs, So I do not know what the problem might be
Can you post these images? Because without seeing the image and your prompt with the settings, I'm unlikely to be able to help you in any way
@prgfrg23 here is it, https://civitai.com/posts/3484592
@prgfrg23 here is no another lora but expressiveH. but less noise becuase negative is more way to anime https://civitai.com/images/14141604
@prgfrg23 no another lora, prompt like 1st image. https://civitai.com/posts/3485435
@infrezz721 I literally cannot reproduce such a result, there is no noise or spots on my images, I have no idea why there is noise and spots on your images
I also can't figure out why the image size is so much different.
from 896x1152 to 1344x1728 (Your image 1600x2057)
@prgfrg23 thanks to response me anyway. I use comfyUI have no this issue. but I still like to using A1111
@prgfrg23 Hey, I found a solution. That all because A1111 python in-built. I try to reinstall external python like an old day and issue is gone. maybe it's sound weird but I just want to let you know.
@infrezz721 I'm glad you found a solution to the problem.
@infrezz721 what do you mean by "like an old day"? I'm having a similar issue with my images being fried and wanted to try the solution you found
@BabyThomDL just delete a1111's system folder and install Python 3.10.6 when you run a1111 don't run on run.bat but run on webui-user.bat in webui folder instead. it's generate slower but make image better quality.
@infrezz721 that's exactly how I'm running it and I get one get image and then all the rest are fried splotches and colors.
ITs great.
95% right eye is bad,when I use it. in 0.5~1 weight.
anyone know why this lora fries my images in the automatic1111 webUI, but works perfectly fine here with the exact same settings?
I use A1111/Forge myself and I don't have any problems, so I do not know why my LoRA can fries your images
Are there any problems with other LoRAs? or just with these?
@prgfrg23 these are the only ones.
however, I noticed that these problems are unique to A1111 as well. I tried forge and they work fine on there
Cuz A1111 is a galactic pain in the ass..? I just spent 8 hrs trying to get the stupid thing to acknowledge that I already had pytorch installed, even after reinstalling pytorch several times and installing the exact version that it told me to..
I've never worked with loras as impressive as these before. Thank you for making them. ♥
Pretty good
A nice selection of styles. I find them especially good for using as accents rather than driving the main visual design.
any idea why it doesnt work on tensorart
Muito bom
good
Great
kool
Loving it!
Do you need to download each individual one or is it all included?
You need to download each one individually (or whichever ones you want, for example I only downloaded the Photo one).
Do i need to use the trigger words for it to work?
Yes, if Trigger Words is there
If there is no Trigger Words then just use LoRA
why?
@javasuperdev297 Hi
These styles replace Smooth/Smooth 2 & Smooth Night
is there any easy way to download them all(yes im aware how many there are, i have the space)
I don't think there's any way to download them all. You don't have to download all the styles! choose the ones you like the most, the size of all 38 files will be 8GB
@prgfrg23 As someone that just tried for 8 hrs to get Automatic1111 working (most of that just trying to get it to realize that I have pytorch installed - failed on that btw..), I don't think I'll ever complain about a UI-based click to download thing ever again..
Great work, thank you :)
Details
Files
Faux Oil Painting Soft Style LoRA_Pony XL v6.safetensors
Mirrors
Faux Oil Painting Soft Style LoRA_Pony XL v6.safetensors
Faux Oil Painting Soft Style LoRA_Pony XL v6.safetensors
Faux Oil Painting Soft Style LoRA_Pony XL v6.safetensors
Faux Oil Painting Soft Style LoRA_Pony XL v6.safetensors
Faux Oil Painting Soft Style LoRA_Pony XL v6.safetensors
FauxOilPaintingSoftStyle LoRAPony.safetensors
Faux Oil Painting Soft Style LoRA_Pony XL v6.safetensors
Faux Oil Painting Soft Style LoRA_Pony XL v6.safetensors
Faux Oil Painting Soft Style LoRA_Pony XL v6.safetensors
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.