This lora aims to replicate the style Meesh uses from 2017 onwards til about early 2025 where his style switches again.
The Krea version is a work in progress. I'm happy with it visually, but it's prompt radiance could use some work.
I'm happy enough with the Anima version where I can call it a V1 and on par with the Illustrious version (IMO). I will definitely keep an eye on ways to improve it further, but for now, I hope you all enjoy!
Description
The first "proper" version on Anima, trained using all the knowledge I've learned over the last month or so.
Trained on 3x the steps as the previous version and as a result, follows the artist's style a lot closer.
Still not perfect, from what I can see in the metadata there's still room for improvement. For now, this is the best I can do.
FAQ
Comments (9)
I miss this style, I don't like his new watercolour/less detail style he has now, it's probably easier and faster but this is the style we know.
Did you train this with the default settings?
Hi there!
For the illustrious version I've unfortunately long lost my training settings. But I trained On Civitai's kohya_ss online trainer with Prodigy + Cosine and I think around 3k steps.
For the Anima version, I used the exact method used in this article https://civitai.red/articles/31678/a-complete-beginners-guide-to-local-anima-lora-training
@Boerboer I Dont plan on training Anima locally, only on Civit. I heard their default settings are kinda bad, and some say their dim and alpha should be smaller; and alpha being half of it
@koopa990 Unfortunately I'm not too well versed on the Civit trainer for Anima. I do agree that dim and alpha should be at least half than default for most cases, but not sure about alpha needing to be half than dim.
What might be worth doing is looking at LoRAs you like and use this tool on them.
https://lora-inspector.rocker.boo/
You can use it to see what steps and settings others use for their training.
Hi everyone, long time no see! I've started experimenting with porting some style LoRAs over to Krea 2. Don't let my contest winner badge fool you, I don't really know what I'm doing yet. Which is why I've labeled this as V0.1. For this version, I recommend using natural language prompts. My next version will likely use tagging instead.
I'm going to continue trying new settings and techniques with other styles, before hopefully revisiting this one for a proper V1. Enjoy in the meanwhile!
New trained lora on tags version when?
@Manly_Men_Enjoyer Hey there. I would guess in about 2 weeks. Just want to keep practicing on a few other LoRAs before focusing on this one again.
@Boerboer Ah thanks for letting us know man!