MugenMaluMix
These checkpoints are merge mixes combined with LoRA finetunes including base models:
Thank you to all the model creators!
For the full merge recipes see the "About this version" section of the model card.
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 prioritize 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
NSFW:
These words help guide the results towards adult 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 ExplicitQuality:
Being mostly a Pony Diffusion base, the score tags still apply, though through successive merges this mix is converging on the more common masterpiece, 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}},
masterpiece, best quality, very aesthetic, absurdresNegative prompt:
score_4, score_3, score_2, score_1, source_pony, source_furry, source_cartoon, 3d, monochrome, greyscale, signature, text, watermarkWorkflow:
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
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
Description
FAQ
Comments (11)
I like your model, can output the character correctly (such as outfit and pose) what I trained. I always use it for lora model testing. Good work
Thank you, I really enjoy seeing the nice results from your LoRA's as well. Nice work!
Can you please allow online generation?
Hello, I've asked to make the model available for online generation - it's not something I can individually make happen as the model uploader sorry.
(Apart from the "Run on Civitai" setting for the model - which I've already set to true)
Good model. I am using v4.2. Comparing to Prefect Pony XL v2 with similar prompts and configurations (I only changed negatives based on this model's recommendation), generally draws hands or feet more correct without specifying negative tags like bad hands. However this seems tending to draw bright and high-contrast arts. I used -0.15 relative brightness in sd-next to try to fix that.
Listens to prompts so pretty easy to use, same as Autismmix. But exhibits varying artstyles, so still a little unpredictable than Prefect. More anime-y out of box than Prefect.
My configuration:
1920*1080 + HiDiffusion
CFG 7 / 25 or 30 steps / euler a
ESRGAN 4x Ultrasharp
-0.15 brightness (relative values)
This is the only model out of everything i used, that takes 2 hours to generate a 512x512 image, what gives????????? i have a 4090 and a ryzen 9, all the models i use take less than a minute, but not this one?
Sorry to hear that, an 512x512 image should only take ~4 seconds to generate with that system.
The previews on this model are generated in Forge, have you tried using generation settings similar to the previews on this model?
@motimalu yes but now whatever prompt i use the image comes out as either a mix of green and black nothingness. I havent had an issue like this before, all the other models i used work fine.
@mochacocah238 I'm not sure what that would be, with Forge you might also try using this: https://github.com/hako-mikan/sd-webui-prevent-artifact (see https://civitai.com/articles/4044)
Though that issue only occurs with certain prompts, and if you can't reproduce the generations from the previews on this model card, it could be something unrelated.
Any plans on bringing this to IL/NAI? I found when using it back in the pony days that it was by far the best for producing monster girls, something that I find Illustrious and NoobAI is lacking atm. Would be great to see a port!
Thank you, yes I could try something - I've also trained a large IL LoRA which includes the datasets used for this mix: https://civitai.com/models/1018079/motimalu-all-rated-illustrious
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.