KitchenSink V1.8 + 1.8L
A SD 2.1 768+ Boosted Artistic General Model

[[ this version is now archived and will not change. Mar 16th 2023 ]]
[[ PLEAE USE VERSION 2 IN MY PROFILE. SFW/NSFW ]]


Added LORA blend. GALLERY for GRIDS!! (Attachment limits sorry! Also civitai renames things I don't know why, on my pc it's named KitchenSinkVer1.8L etc. I suggest renaming to keep nice and clean after downloading)

[ This model is a merge of all 2.1 based ~ 768+ models because I got sick of swapping between models and 512px is too low really. ]

• What the model differences are:

  1.8 = (the sloppy fun drunk) broad artistic vivid. Not scared to spill the wine while tipsy painting. Oh I made a Salvator Mundi? Cool.

  Example: The flower rose prompts. Vivid & vibrant.

  1.8L ( for Lora blended in) = Less drunk with a sharpened crayon. Dinosaurs? ya gotchya, I won't make them as gross (Thanks dino lora's!) BUT it is much less vivid and punchy vibrant.
  
  

• Base 2.1 !!! > Don't forget the included YAML! (place in same folder as model. rename to same exact name as model! IE: KitchenSinkVer1.8 )

  [FYI You still need a yaml for 2.1 stuff - I don't know why some people say you don't anymore. For those reading it's less the auto11 not needing it and more that it is starting to give errors when merging/pruning/etc if you do not include your custom YAML file.
  -> On this model please use my included sink yaml merci!]
  

• Resolutions 768+ (768x768, 768x1024, 1024x768 ++ etc)
  

• No clip nonsense, no vae nonsense, bog standard SD setup, simple to complex prompting as you wish, negatives not really required (fyi: the more negatives you enter without being excessive the better SD 2.1 preforms by the way!).

  IE: type cartoon if you want cartoon, realistic for real etc. call to it!! It has everything built in!! A few "realistic" keywords ? start with " masterpiece photo of... " and toss in "3d render, cartoon" into the negatives. =)
  

• Sample gallery is random raw internet prompts on a fresh install - first try, no redo's, no backsies, sequentially numbered!, non cherry picked images. ( Fair is fair common! haha)

• LOW Step count needed. Fantastic results in the 20-25 step range!

• XYZ cat grid included to show 17,22,30 steps on all samplers (Standard CFG 7) "a cat lounging in the grass, basking in the morning sunshine" Note: some samplers do not work on (my) old gpu hence the few trash outputs. Apologies for those bits as it will probably work for you.)

-->>[ SFW/NSFW - ??? totally untested (to the level some people want). As it's a total merge, anything is possible so don't be mad if something slides in - just add a negative word prompt to compensate. See merging info below. ]
** The model "seems" and feels generally safe for work. Just..yeah! unknown! =)

WHAT IT LOVES AND HATES

It loves >??> Flowers certainly so far! (Bloody brilliant!) It also seems to understands a wider artistic range of broad subjects and prompts. only needing light negative prompting is a bonus as well. It also puts an interesting sort of "oil paint realistic edge" ?! as it's own artistic style to things.

It's good at >++> generally everything really ~ but more artistically than realistic from all the mixing. I randomized prompts from every site to make the gallery here and i find the results better in most cases when VS other general non purposely tuned model. (it seems to listen better)

It's bad at >--> For one it lost the face/hands of the realistic models being diluted and has trouble with them in general now. And Secondly, It takes a few tweaks here and there. Meaning you might need to bump the CFG or steps depending on your purpose. Bit of fiddling is required depending on your purpose. Think ..tuning a guitar! see next point.

-->> Not a bad point but really good to know: It's also a bit of a handful as larger models generally are, meaning sometimes you just need to run the same prompt a few times as it's got an eclectic mix of seeds deviating from the standard.

I found that you cannot rely on a single result at times as it just hit a "bad" seed for your scenario. It's a hugely packed model that has an enormous seed resource pool built in! Making it a handful... completing the irritating loop haha. See the mixing below)

Copyright >>> I personally don't give 2 hoots what you do. Know that this has every model from every human that made one in 768+ rez until now mixed in. If it had a yaml it's in here. So, like, you probably gotta kinda sorta maybe follow their rules as it was originally their models. =)

** To be clear of what is in this model merge. All working non biased overcooked models in 2.1, with working yaml file, in 768+ resolution, from aug '22 to feb '23 are merged in. (Approx 700GB of models total) - which is why the pruned versions cannot be lowered past the 5GB size. I'm sorry it's as small as I can crush it down without clipping data. The version 1 description below has the long tedious formula I use.


Post your comments and quibbles below.

Please enjoy responsibly!

BarQ!

> Hit the buymeacoffee link under my profile icon to help with the next R&D project!

> Visit my artstation link to view custom built offline AI creations as well!