Click here to join the discord and share your images ♥
♥ a model for gooners ♥
Mugen is a stylized anime/realistic model for SD 1.5 and SDXL.
Prompting compared to PDXL v6 is generally easier, in turn simpler prompts generate better quality images.
All sample images were generated using ADetailer on the face and hands.
(08/07/24) v3.1 released. If you were here for v3-q, it was quite sloppy and had trouble following prompts, but was great at following specific styles. I hope to remediate this with v3.1
Please check the About this version section for model specific info.
Mugen 1.5:
For Mugen v1-3, generally just use the guidelines given in v3.
Mugen XL:
For Mugen XL v1-3.1, prompting isn't much different across versions, but experimenting is always recommended.
Try adding "3d, blender, " to your negative prompt near the beginning (preferably after your score tags) to avoid that style if you don't want it. XL v1 and v2 heavily default to that style of image. While it's not as apparent in XL v3.1 with proper style tags in your positive prompt (or LORAs), it's still frequent enough to be worth prompting out.
♥
Description
Share your generations with us in our Discord server!
Mugen ∞ v3 is an NSFW-prone, stylized/semi-realistic model.
All sample images were generated with very simple prompts (which can be found below), and a variety of samplers, using FastNegativeV2, and the following parameters:
Clip skip: 2
Steps: 32
Resolution: 512x768
Hires-fix Upscale: 1.5x
Hires steps: 12-15
Upscaler: 4x_NMKD-Siax_200k
Denoising Strength: 0.5
CFG Scale: 7
ADetailer inpainting denoise: 0.3
positive prompt:
1girl, large breasts, slim waist, thick thighs, <insert short auto-generated prompt for clothing>, landscape
or for the more stylized sample images;
anime, 1girl, large breasts, slim waist, thick thighs, <insert short auto-generated prompt for clothing>, landscape
negative prompt:
FastNegativeV2