A LoRA trained on over 1400 popular artists from e621, meant to be able to add some control over the style of Pony Diffusion V6 XL, in a similar way to how e621 base models can be controlled.
Use "by <artist>", check the PDartistsV2.txt file (in "training data", also usable as wildcard)
v2.0 changes (compared to v1.0):
1436 artists (from 224 originally)
bigger dataset (350k+ images, from 80k+ originally)
better balanced dataset (same amount of training steps for each artist)
Description
FAQ
Comments (16)
Excellent work.
And here I felt like I was being excessive when I scraped 10k images from e621 yesterday.
Hi, I'm trying to restore more artists to pony as well. Can you share your training settings?
dimension 128 (probably slightly overkill for this, could halve this), alpha 64, unet lr 1e-4, te lr 5e-5, optimizer AdamW (betas=0.9,0.99 weight_decay=0.1), 5% caption dropout
That this can be used to create an XL model that is a merge between YiffyMix and Pony Diffusion.
You legend, you fixed the model.
is this for 1.5?
No, it's for Pony Diffusion V6 XL https://civitai.com/models/257749/pony-diffusion-v6-xl
and pony diffusion xl v6 is given the name, based on SDXL, sdxl lora's can work on pony sdxl models but pony sdxl loras cant really work on regular sdxl.
All 100+ posts artists from e621 lora when?
Soon
@Tails8521 beg for spaal to be there. He have 98 posts 😭
Your pfp, is that winter wyvern's ult(winter's curse)?
@idiotlol32 Yea
Hi, question please, what's the correct weight for this lora? Has triggers? Ty.
Anywhere from 0.3 to 1.0 depending on how much you want it to influence the style I guess, the triggers are the artist names
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.