Zackary911 style for PonyXL.
Illustrious v1
(Dec 29, 2024)
Pretty much copied everything from Pony v2
Trigger: z9style
PonyXL v2
(Dec 29, 2024)
Cleaned up the dataset by removing signatures and borders. It looks a bit different than v1, no idea why. I tried a dozen times to get it to look better and eventually settled for this. I think Illustrious is spoiling me, heh. Just play around with different checkpoints if you want to stick with ponyXL.
Also added a trigger word: z9style
PonyXL v1
(Apr 12, 2024)
I had a bit of trouble with this one, but it's pretty close in most cases. Works best with furry concepts. Very good at reproducing the artist signature unfortunately. I may fix it when/if I do a v2, but in the mean time you can inpaint it out.
Preview images were generated in pairs with Pony Diffusion v6 and AutismMix-confetti respectively.
No trigger required
Weight: 1.0 - 1.3
Description
Dataset and settings copied over from Pony v2
FAQ
Comments (9)
Illustrious's ability to understand and integrate things correctly has me a bit spoiled as well lol.
Indeed. as long as the LORA itself isn't bad, ILL seems to be just shy of perfect.
Enjoy the 1k tip <3
An excellent Lycoris. Do you plan to expand upon it? You said you pretty much copied everything over to illustrious from the Pony model, but the illustrious is optimized differently. So maybe it could be more effective with the things we've learned about Illustrious in the past month and a half? Maybe the upcoming illustrious 1.0 is a reason is a good excuse to revisit this project?
Just a thought! :D
Oh, I haven't heard anything about an illustrious update. But I guess it depends how much of an improvement it is. Right now, base illustrious 0.1 gens blow ass. All I can get out of it is low quality deviantart looking stuff. The merges and checkpoints are what's carrying it. So we'll see if the next version is good out of the box, or if we have to wait for updated checkpoints too.
This lyco's been finicky. Like right now, the shading looks alight, but it mangles penises. No idea why. I think I have to be more selective with the dataset, regardless of base model. I'll have to come back to this one for sure.
@H4fun91 Yeah, I agree. The merges and checkpoints are what's carrying it, at least for furry stuff in my experience. I have had experiences with mangled peens but only when my prompts are very specific and detailed. And simply tweaking the prompt can fix it after enough trial and error. If you want an example, if you're working with a first person perspective shot with a character who has large peen and a womb tattoo, sometimes it displays both correctly with the rod being in front of the tat, but sometimes it'll break or even warp into a blog of flesh just to make the tat visible. Sometimes I'd find something good, tweak one thing, and it all comes crumbling down. (Mind you, this was while using the same personalmerge used in your red panda example image, while it is WAY better than the base 0.1 model it's obviously not perfect haha)
But yeah! The new 1.0 version becomes free to the public in little over 5 days from now, and apparently it still can't draw the correct number of fingers even with a simple prompt and reasonable settings. It'll definitely be a "wait for checkpoints/merges to come out" type deal.
Update: From what I see, It's way better at drawing Anthros then 0.1 but still nowhere near the level of merges and checkpoints. Maybe that means merges built off it will be better? Only time will tell.
@Wallace_Tanaka Yeah, just looking at the gallery there it doesn't seem like anything special right now. But v0.1 is way better than ponyXL in terms of anatomy and prompt recognition, so they're doing something right at least. I'll probably have to play around with it. See if simply training with it yields benefits?
Awesome style!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.