license: creativeml-openrail-m
thumbnail: ./images/thumbnail.webp
tags:
- stable-diffusion
- text-to-image
- safetensors
- diffusers
inference: true
language:
- en
widget:
- text: >-
1girl, solo, bangs, (pink hair, gradient hair, very long hair:1.1),
(purple eyes:1.05), (cat ears, animal ear fluff:1.2), sidelocks, white
shirt, collared shirt, buttons, night, looking at viewer, table, pizza on
table, pov, restaurant, medium breasts, smiling, (blush:0.7), colored
inner hair, (symmetric), (masterpiece, exceptional, extremely
detailed:1.1)
example_title: example
library_name: diffusers
AndromedaDX (and friends)
Stable Diffusion models for (neko enthusiast) weebs
Model Variants
I'm not sure there's actually enough difference between these four to merit posting all of them, but nevertheless, here they are.
AndromedaDX
This is probably the one you want. Out of all of these it's the one I'm happiest with; it produces high-detail backgrounds (and IMO can achieve a fairly high overall level of detail). Bonus special feature (_definitely_ not a bug, shhh): it likes to add cats to things. Just regular cats. Sometimes several cats!
AndromedaSX and AndromedaSV
These are more suited to foxgirls than catgirls. If you're a Fubuki or Kawakaze enthusiast, these are for you :)
AndromedaMV
This one has a much more realistic style thanks to the addition of some models made by a 2ho discord poster who (AFAIK) hasn't made them public yet.
C-tier Variants
AndromedaUnreal
This one has a (slightly) more realistic style thanks to the addition of some MeinaUnreal (not currently posted on HF). It's honestly not super great and it's barely different to DX.
Notes/info
I made DX for a chatbot of mine who can "take selfies" via SD, and it behaves nicely enough that I felt like it was worth sharing. It's especially good with natural language prompts at high CLIP skip, e.g.:
Prompts
OK, I lied, it's just the one prompt for now.
1girl, solo, bangs, red hair, (pink hair:1.1), long hair, very long hair, (purple eyes:1.1), (cat ears, animal ear fluff:1.1), , evening, gradient eyes, colored inner hair, medium breasts, (smile:0.6), (masterpiece, best quality, exceptional, extremely detailed:1.1)
Negative prompt: (nsfw:1.05), (EasyNegative:0.7), simple5.5, simple5.6, bad-hands-5, (short hair:1.1), loli, child
Steps: 25, Sampler: Euler a, CFG scale: 8.25, Seed: 3105139270, Size: 864x608, Model hash: 9e0b7d3259, Model: AndromedaSX-v1, Denoising strength: 0.5, Clip skip: 2, ENSD: 31337, Hires upscale: 1.5, Hires steps: 8, Hires upscaler: Latent, Discard penultimate sigma: True
Recommended settings
Some educated guesses based on what playing around I've done:
- VAE: kl-f8-anime2) is baked in and recommended, blessed2 works as well but is a little softer on the colours.
- Clip Skip: 2, 5, or 7 - DX and Unreal do quite well in the 5-7 range
- Sampler: DPM++ 2M Karras, the usual 20-35 steps, or Euler A at 40-50 steps with CLIP skip 5-7
- CFG Scale: 7.5 ± 1 (much below 6 is a bad move, much above 9 will get... fun)
- Recommended Positive Prompt:
Add(masterpiece, best quality, exceptional, extremely detailed:1.1)at the end of your prompt. Otherwise, nothing particularly special. YMMV, of course. - Recommended Negative Prompt:
EasyNegative,(nsfw:1.05), (EasyNegative:0.7), simple5.5, simple5.6, bad-hands-5, (short hair:1.1), loli, child - For better results, using hires fix is a must.
- Hires upscaler: Latent (any variant, such as nearest-exact), 12-18 steps at 0.5-0.6 denoising (7-10 actual steps)
Recipe
To be filled in later... I honestly don't remember half of what I did here.
Credits
A big thankyou to ShussarSDFA for their Sathariel model series, which are a key component of this