This is a merge between my own drawing style + three other artists (One of them being Ratatatat74 which seems to be overpowering the most).
So, I called it the Ekune Style LoRA. Though, unfortunately it seems the other styles overpowered my own and only small glimpses appear every once and while but I still like the results. I trained it using 40 pictures varying from when I first started up until last year to make a LoRA of my own art style - It felt lack luster, so I fused it with three other LoRA for better coloring, lighting, etc.
This is an OC of mines to show you my art: Here
It only has four pictures but it shows how I evolved starting from making 2D Chibi characters up to the latest picture.
I suggest using this LoRA at 0.8-1.0 weight for best results.
All examples images were generated on AOM2 and Anything V3 VAE.
All pictures were upscaled using these settings:
Upscaler: Fatal-Anime 4x
Hires. Steps: 25
Denoising Strength: 0.5
Upscale by 1.5 (768x768)
Embeds and LoRAs used
Bad Prompt - Bea - Misty - Ryuuko - Giovanna - Yohana (OC)
Follow me on Twitter for updates- https://twitter.com/TrueHeroBang
Support - https://ko-fi.com/ekune
Description
FAQ
Comments (12)
Hey! your next Artstyle Lora since you're planning one should really be in the style of Mamimi / Mamamimi, He has such a unique but amazing style.
I'll check it out tonight.
Very cool. Do you feel like it captures your style well? Looking at the sample images you linked to, I can certainly see the resemblance, but I’d imagine it’s much more personal when it’s your own work.
It does capture my style well - It also improves on it greatly because of the other artists merged with it and because AI will do AI things. I'm glad that I tried it. I'm happy but it also makes me a little sad as well, ya know.
@ekune do you see yourself continuing to develop your skills as a traditional artist? Or do you plan on using AI for everything going forward?
@justmaier I think the plan is to continue improving traditionally while using AI to aid me either by helping me with composition, shading, or coloring which I believe are some of small weak points that I have. AI just helps bringing my ideas to life faster than before.
I love the super clean color palettes and line work of both your original artwork and the generated LORA ouput.
I noticed your included a "pixel-like" sprite in your dataset. I wonder with Control Net + a Pixel LORA / Prompt could you maybe use this to start generating animated sprite sheets without having to draw every-single-frame.
I will definitely check this out when I have free time later tonight.
[Will Post a Review after trying it out]
Only used three sprites in my dataset - so I don't know how well that would work but In the future, I could most certainly make a dataset with all of my sprites to see how it goes.
Have you tried this extension though: https://github.com/AUTOMATIC1111/stable-diffusion-webui-pixelization
This is an example of something I made with it - It's a gif, mainly the mouth moving though! - https://imgur.com/ToeA2il
You could even use https://civitai.com/models/10706/luisap-pixel-art-lora @ 0.6-0.7 weight with my LoRA to create pixelated pictures if you're going for that specific style.
Example: https://i.imgur.com/WlJpyUM.png
@ekune
Thanks I'll try out that extension.
Didn't have much luck with the luisap lora, but I'll give it another shot with that setting.
(Your posted examples look fantastic)
--
I also found:
https://www.pixelicious.xyz/
But the underlying model is not release and I prefer to run locally.
(It does a nice job pixelating & background removal - but you only get X amount before you need to pay. )
--
What I've been looking for though is what your doing here, Nice clean colour palettes, with only a few shades for shadow / lighting passes but nothing excessive - and thick lines.
This is great, please make more of these style mix Loras.
Would you mind sharing how you merge Loras? It really slows down my gens with the 0.5 0.5 0.2 0.2 0.5 mess I'm doing right now
I used https://github.com/hako-mikan/sd-webui-supermerger this extension.
Click on SuperMerger -> Lora -> Choose the LoRAs you want to merge at the bottom -> Click on Merge LoRA. That's all.
is old but good one style mix, any chance for sdxl illus version?