When you say you are thinking of redoing the early game do you mean the very start before the different routes kick in?
Sorry! Totally forgot to reply
Yeah, I’m not really happy with the early-game pacing or how it's handled technically.
I’d really like to start things off with better renders and a better look at where Mila and Paul are emotionally at the beginning.
No idea when I’ll get around to it, but I definitely want to fix it someday
Hey I bet plenty of people asked you that already but you stated that you use Pony Diff to create your images. Do you work a lot in Img2Img tabs to create your art ? + What UI do you use? A1111 ? Forge? What is your go to Lora that you use in most of your pics? I myself started creating some stuff and have plenty of fun. Was just wondering.
Fyi Im using the NoConceptArt, Incasestyle, easynegative, princessXL ect. Have a great day!
I’ve got a small guide in Discord with a sample prompt — it’s public and available to everyone.
It’s a bit outdated by now, though. I just haven’t had the time to write a new one for every update to my workflow.
I use several models — pretty much moved away from pure Pony. These days my go-to is the
Illustrious family.
I use a mix:
realistic,
cyberrealistic, and
mix at different stages of the process.
For refiners I usually go with
Ponyfaetality on very low denoise, and I switch around 0.7–0.9 depending on the base model.
I also use a
lot of Photoshop, references, and third-party generation tools (like ChatGPT-style stuff — though honestly, the censorship there is straight-up bonkers.
Sometimes it won’t even let me generate completely SFW stuff for god knows what reason. But hey, at least it helps with building complex compositions and backgrounds from the right angles. So that’s something.)
As for UI — I still prefer
A1111, even though people keep saying it’s outdated.
For me, nothing beats its
inpaint workflow — everything else is clunky as hell or messes up the style.
I use the default vanilla UI, with just one extension for setting custom aspect ratios.
I’ve got links to some of my LoRAs in Discord — the standard pony-style one, plus a custom
Mila LoRA.
I also trained LoRAs for
Paul,
Kiki, and
Liz, but I haven’t made those public yet.
Here’s my basic workflow: – First I sketch out a draft using references or draw one manually.
– Then I run
img2img at 0.2–0.3 denoise.
– Fix stuff in Photoshop.
– Run another
img2img pass at the same denoise.
– Repeat that cycle a few times until I’m happy with the composition and anatomy.
– Then I use
inpaint to fix limbs, clothes, faces, and smaller details.
All in all, it usually takes around
150–200 generations per image, give or take.
Sometimes it goes faster if the reference is spot on — usually
composition is the most time-consuming part.
But sometimes I spend
hours just trying to get the right emotion, and that turns into 100–200 inpaint passes
just for the face, until it finally hits the right note