Style lora based on LilAndy/Andava.
I removed all the watermarks manually from 100+ images before training, so this wont have the issue other models of the same artist have.
Depending on the model you might need to up the strength to higher than 1 numbers or use a higher cfg for the style to show properly.
Some IL models already recognize the artist artstyle, so puting "Andava" on your prompt will give you a really accurate version of the style, but will also bring back the watermark issue, so use a low strength like (andava:0.5)
I would appreaciate if you could like and post your images, as that help me get more buzz to keep making more models.
Description
FAQ
Comments (10)
Almost every image on the OG post got nuked by moderation. So be wary about that when making images with this model.
If you know the artist you know what I'm talking about.
This one is also well made lora for the artist's style.
I like it.
Do you recommend any tutorials on how you achieved the results you did? This works phenomenally.
If we're talking about lora training, I just used mostly default settings on the CivitAi lora training. I think I added more steps and took the model that worked best from the bunch. Even used the auto tagging of the site.
The only actual recommendation I can give is to go through the images manually, keep only the better ones, and edit out things like signatures and text manually (I use the windows generative erase tool, it works wonders most of the time)
@Liviel Yeah, that's what I was referring to. Thank you!
Awesome style!
This model can be good finisher/polish for some combination of style. Really well done
Very good job. It's very faithful to the artist. I'd love a flux version. If you'd be willing to give me your training data for this model, I'd do the training for flux.
Очень красивый стиль но модерация херовая, почти 90% изображений с этим стилем не пропускает... Как и Dagasi. кто знает тот поймет.
Word brother. Let's hope it gets better or another site allows it.