Balanced - All-In-One - photoreal, anime, pixelart, SFW/NSFW, styles, furry, ... (vae-baked)版本v1.0 (ID: 772228)

Quick summary:

SFW/NSFW (you may need to mention clothing in the prompt)

The settings I use (not quite the same as "suggested settings") :

15-20 steps

896x1152

CFG 2.5 - 6 (3=default)

DPM++ 3M SDE, Karras or Polyexponential

HiRes Fix, 15-20 steps, resize x1.5, denoise 0.5

Works great with all the LORAs I've tried so far.

Prompting: <style: polaroid/pixelart/etc> <subject> <background> <extra>

-"(pixel-art:1.4), square pixels, man in mechsuit standing atop a mountain range, clouds, wind, rain, storm, ..."

-"source_anime, source_furry, score_9, vivid watercolor, ..."

-"polaroid photo..."

Obligatory rambling summary:

I dwell in the lawless garbage-wastes beneath CivitAI's core, emerging by day to search the marketplace for pretty gems which I steal and crush into dust so that I may roll them into my ever-changing chaotic ball of dough. Basically, if it doesn't look like shit, I download it and I throw it in. If the result is an improvement, I keep it. If not, I throw it out. I merge the best results together until I get something new. That is my process. Usually, I keep the results to myself. Today, you - the Hobbitses - may have a copy of My Precious.

I merge and I merge and I merge. It is one of my hobbies by now. I merge and re-run old test-prompts on those merges more often than I actually sit down to 'get creative with prompting' anymore. This model in particular seems to be pretty damn good at everything I throw at it.

This particular merge was somewhat of an accident. I wasn't aiming to make an all-in-one merge, I was only trying to 'bring in more/sharper photorealism', but then I started testing it. And I kept testing it. And I started writing new prompts, and got good results for all of them. So I knew it was good.

My primary objectives are 1) coherent details and 2) variety of style/concepts. The most recent merges include NewReality and NatVis (just a little bit of natvis because it made the model all fuzzy), but I've been kneading this ball of dough together since SDXL was released.

I've recently merged in some LORAs, like add-detail, a hands-improver, and 'Pony Custom styles'.

There has been a lot of Pony folded into it, though NSFW has not been my goal. I only threw Pony models in to see if it would shake up the art style - which it did - but as SDXL models get better in general, it seems that the best models/merges also get better at NSFW output.

We all know the drill by now: it likes to draw women, and naked women, so if it gives you nudity - which it will at some point - just add clothes to the prompt.

It also uses the Pony VAE, baked-in.

Plugin Shout-Outs:

Perturbed-Attention Guidance and ADetailer can be very cool to play with, though I almost never use them.

描述:

训练词语: