MugenMaluMix

These checkpoints are merge mixes combined with LoRA finetunes including base models:

For the full merge recipes see articles:

The intention of these Checkpoint+LoRA merges is to create a model capable of producing highly aesthetic Game Art, Anime, and Character subjects in a Fantasy style.

Quality goals are to prioritise better anatomy (hands, feet particularly) and maintain the base models responsiveness to booru tags.

If you have any issues running this model, its recommended to use the Forge webui: https://github.com/lllyasviel/stable-diffusion-webui-forge

As well as this extension if you get noise outputs:
https://github.com/hako-mikan/sd-webui-prevent-artifact

Special tags:

The models are receptive to triggers of LoRA included in the merge, see the full list of LoRA's in each of the merge recipe articles.

Concepts:

nekomata, cat girl
kittunemimi, fox girl
usausamimi, rabbit girl
kuronagamimi, dark elf
dragon girl
slime girl
fangs out
oni

dvr-trd, ruby
green emerald style, emerald
iridescent opal style, opal
ral-obsdn, obsidian

Characters:

olohpnxl, holo, spice and wolf

nereirfpnxl, frieren
nurefpnxl, fern \(sousou no frieren\)
lebupnxl, ubel \(sousou no frieren\)

eihplyspnxl, sylphiette
siretaryergpnxl, eris greyrat
yxorpnxl, roxy migurdia
enialsihgpnxl, ghislaine dedoldia
rudeus greyrat
elinalise dragonroad
pursena adoldia
rinia dedoldia

Copyrights:

nier riencarnation
knights of glory
nier automata
bravely default
final fantasy tactics
granblue
fire emblem heroes

Artists:

happoubi jin
lack
kintadistortion
gusha_s
yunimaru

750+ artist styles from 4th Tail checkpoint merge (with a less predictable effect)

NSFW:

These words help guide the results towards ***** content, as pony diffusion and related checkpoints are heavily used in these checkpoints, it's best to use these if SFW is desired.

Use in positive/negative prompts as needed:

safe	            General
sensitive	        Sensitive
nsfw	            Questionable
explicit, nsfw	    Explicit

Quality:

Being mostly a Pony Diffusion base, the score tags still apply, though through successive merges this mix is converging on the more common masterpeice, best quality, very aesthetic, absurdres tagging standard.

Recommended prompt structure:

To guide the model towards generating high-aesthetic anime style images:

Positive prompt:

score_9, score_8_up, score_7_up, score_6_up, source anime,
{{artist}} {{character}} {{copywrite}} {{tags}},
masterpeice, best quality, very aesthetic, absurdres

Negative prompt:

score_4, score_3, score_2, score_1, source_pony, source_furry, source_cartoon, 3d, monochrome, greyscale, signature, text, watermark

Workflow:

Adetailer extension is used in some preview images, and is highly recommended to automatically segment and upscale faces/hands/bodies as needed:

--> https://github.com/Bing-su/adetailer

Merge Recipe:

To test out/reuse the merge recipes from v4.2 ComfyUI and several DARE extensions are required.

ComfyUI JSON for the DARE base model merge and DARE LoRA finetuning have been added to the Model Training Files.

⚠️These are quite messy, and will require significant hardware to run without modification - I used a machine with 64gb of system ram.

License

As the license of PonyDiffusion is more strict, the model will follow the same modified version, quoting the following from the PonyDiffusion model:

The following modifications have been added to Fair AI Public License:

You are not permitted to run inference of this model on websites or applications allowing any form of monetization (paid inference, faster tiers, etc.). This applies to any derivative models or model merges.

Explicit permission for commercial inference has been granted to CivitAi and Hugging Face