💰Follow my Patreon -> CLICK ME
📧Join my Discord -> CLICK ME
⭐If you like my work, leave a review. I really appreciate them!☑️
I have a new DoRA version, please use that one for best results ->CLICK ME!
Sleepygimp Style Lora. I recommend using 0.8~1.0 strength.
Trigger word: sleepygimp
Please look at example images for help
I use additional tools for my generations so my images may be different from yours. If you want to know what I use, please look at my guide.
My Socials | Support me on Patreon or Ko-fi (If you want)
Description
Initial Release
FAQ
Comments (9)
I wondered when this art's style would show up.
Will try it out!!
Thank you for sharing / doing this.
Hope you enjoy it!
awesome looks great kinda like Scott campbell art! (I would love a model with his style but this looks great too! thanks)
@Cube3D I have a request form. If you fill it out and follow the rules for it I can make the style for you. I do these completely free, I just only ask for datasets on a google drive.
@guy907223982 Awesome!! gonna try! Thanks for the opportunity!
Really seems like it only works with ComicBabes..
Interesting cause you trained it over AnyLora.. I tried it with several checkpoints ranging in styles / line thickness and the art style really doesn't come through at all..
Maybe it needs to cook longer? Not sure..
It's a combination of training with my older training method, and many checkpoints being trained off anime. I have found a couple models when I originally tested worked, but I found comic babes is the best with it's consistency. Really we just need someone to make a model that is a mix of a more "western art style" that works with training. I have seen a couple of people try to accomplish this, but it seems to be a very hard task and I have no idea what the complexities of that are tbh.
@guy907223982 I know the mantra for training styles is to lower the learning rate.. but I find that actually washes the style out quite a bit.
I've been having more luck on style training but using the standard learning rate but then also getting a large dataset and captioning it well.
Then for models that are similar art style to the style you want - it's very accurate.. and for models that aren't, aka photorealistic for a artist style.. you get a mixture of "oh that's kinda neat" to "dear god what abomination have I created"
I'm still learning how best to train and setup datasets myself though.
@DreamExplorer yeah, its not a simple matter. It's very hard to get these art styles right, it's not as simple as a character where you just load your preset settings and let it run. Doing styles if you want to get them perfect will take a lot of testing, sometimes it works fine right out the gate. Other times, it just looks like a general anime model. Sometimes SD can't recognize stuff like undertones or lighting as it try's to overcorrect them when generating. Not really sure how you fix that currently.